# correlation coefficient

107 results back to index

Statistics in a Nutshell by Sarah Boslaugh

The different t-tests and their uses Chapter 7. The Pearson Correlation Coefficient The Pearson correlation coefficient is a measure of linear association between two interval- or ratio-level variables. Although there are other types of correlation (several are discussed in Chapter 5, including the Spearman rank-order correlation coefficient), the Pearson correlation coefficient is the most common, and often the label “Pearson” is dropped, and we simply speak of “correlation” or “the correlation coefficient.” Unless otherwise specified in this book, “correlation” means the Pearson correlation coefficient. Correlations are often computed during the exploratory stage of a research project to see what kinds of relationships the different continuous variables have with each other, and often scatterplots (discussed in Chapter 4) are created to examine these relationships graphically.

Calculating Cramer’s V The Point-Biserial Correlation Coefficient The point-biserial correlation coefficient is a measure of association between a dichotomous variable and a continuous variable. Mathematically, it is equivalent to the Pearson correlation coefficient (discussed in detail in Chapter 7), but because one of the variables is dichotomous, a different formula can be used to calculate it. Suppose we are interested in the strength of association between gender (dichotomous) and adult height (continuous). The point-biserial correlation is symmetric, like the Pearson correlation coefficient, but for ease of notation we designate height as X and gender as Y and code Y so 0 = males and 1 = females.

Correlation Statistics for Categorical Data The most common measure of association for two variables, Pearson’s correlation coefficient (discussed in Chapter 7) requires variables measured on at least the interval level. However, several measures of association have been developed for categorical and ordinal data, and they are interpreted similarly to the Pearson correlation coefficient. These measures are often produced using a statistical software package or an online calculator, although they can also be calculated by hand. As with Pearson’s correlation coefficient, the correlation statistics discussed in this section are measures of association only, and statements about causality cannot be supported by a correlation coefficient alone.

Analysis of Financial Time Series by Ruey S. Tsay

The concurrent, or lag-zero, cross-correlation matrix of rt is defined as ρ0 ≡ [ρi j (0)] = D−1 Γ0 D−1 . More specifically, the (i, j)th element of ρ0 is Cov(rit , r jt ) i j (0) ρi j (0) = = , std(r ii (0) j j (0) it )std(r jt ) which is the correlation coefficient between rit and r jt . In a time series analysis, such a correlation coefficient is referred to as a concurrent, or contemporaneous, CROSS - CORRELATION 301 correlation coefficient because it is the correlation of the two series at time t. It is easy to see that ρi j (0) = ρ ji (0), −1 ≤ ρi j (0) ≤ 1, and ρii (0) = 1 for 1 ≤ i, j ≤ k. Thus, ρ(0) is a symmetric matrix with unit diagonal elements.

To overcome this difficulty, we use the simplifying notation of Tiao and Box (1981) and define a simplified cross-correlation matrix consisting of three symbols “+,” “−,” and “.,” where 1. “+” means√that the corresponding correlation coefficient is greater than or equal to 2/ T , 2. “−” means √ that the corresponding correlation coefficient is less than or equal to −2/ T , and √ 3. “.” means √ that the corresponding correlation coefficient is between −2/ T and 2/ T , √ where 1/ T is the asymptotic 5% critical value of the sample correlation under the assumption that rt is a white noise series. CROSS - CORRELATION 305 Table 8.1.

This bivariate GARCH(1, 1) model shows a feedback relationship between the volatilities of the two monthly log returns. 9.2.2 Time-Varying Correlation Models A major drawback of the constant-correlation volatility models is that the correlation coefficient tends to change over time in a real application. Consider the monthly log returns of IBM stock and the S&P 500 index used in Example 9.2. It is hard to justify that the S&P 500 index return, which is a weighted average, can maintain a constant correlation coefficient with IBM return over the past 70 years. Figure 9.5 shows the sample correlation coefficient between the two monthly log return series using a moving window of 120 observations (i.e., 10 years). The correlation changes over time and appears to be decreasing in recent years.

pages: 923 words: 163,556

Advanced Stochastic Models, Risk Assessment, and Portfolio Optimization: The Ideal Risk, Uncertainty, and Performance Measures by Frank J. Fabozzi

The resulting measure is the Pearson correlation coefficient or simply the correlation coefficient defined by(5.19) where the covariance is divided by the product of the standard deviations of x and y. By definition, rx,y ∈[−1,1] for any bivariate quantitative data. Hence, we can compare different data with respect to the correlation coefficient equation (5.19). Generally, we make the following distinctionrx,y < 0 Negative correlation rx,y = 0 No correlation rx,y > 0 Positive correlation to indicate the possible direction of joint behavior. In contrast to the covariance, the correlation coefficient is invariant with respect to linear transformation.

When its value is negative, we say that the random variables X and Y are negatively correlated, while they are positively correlated in the case of a positive correlation coefficient. When the correlation is zero, due to a zero covariance, we refer to X and Y as uncorrelated. We summarize this below: −1 ≤ ρX ,Y ≤ 1 −1≤ ρX,Y < 0 X and Y negatively correlated ρX,Y = 0 X and Y uncorrelated 0 < ρX ,Y ≤ 1 X and Y positively correlated As with the covariances of a k-dimensional random vector, we list the correlation coefficients of all pairwise combinations of the k components in a k-by-k matrix This matrix, referred to as the correlation coefficient matrix and denoted by Γ, is also symmetric since the correlation coefficients are symmetric.

That is, the correlation coefficient161 of two random variables X and Y, denoted by ρX,Y is defined as(14.22) We expressed the standard deviations as the square roots of the respective variances and Note that the correlation coefficient is equal to one, that is, ρX,X = 1, for the correlation between the random variable X with itself. This can be seen from (14.22) by inserting for the covariance in the numerator, and having , in the denominator. Moreover, the correlation coefficient is symmetric. This is due to definition (14.22) and the fact that the covariance is symmetric. The correlation coefficient given by (14.22) can take on real values in the range of -1 and 1 only. When its value is negative, we say that the random variables X and Y are negatively correlated, while they are positively correlated in the case of a positive correlation coefficient.

The Intelligent Asset Allocator: How to Build Your Portfolio to Maximize Returns and Minimize Risk by William J. Bernstein

Lastly, Figure 3-5 plots two very poorly correlated assets (correlation coefficient of .068): Japanese small stocks and REITs. This plot is a “scattergram” with no discernable pattern. A good or bad result for one of these assets tells us nothing about the result for the other. Why is this so important? As already discussed the most diversification benefit is obtained from uncorrelated assets. The above Math Details: How to Calculate a Correlation Coefficient In this book’s previous versions, I included a section on the manual calculation of the correlation coefficient. In the personal computer age, this is an exercise in masochism.The easiest way to do this is with a spreadsheet.

A three-person office has three interpersonal relationships; a 10-person office has 45 relationships.) Real assets are almost always imperfectly correlated. In other words, an above-average return in one is somewhat more likely to be associated with an above-average return in the other. The degree of correlation is expressed by a correlation coefficient. This value ranges from ⫺1 to ⫹1. Perfectly correlated assets have a correlation coefficient of ⫹1, and uncorrelated assets have a coefficient of 0. Perfectly inversely (or negatively) correlated assets have a coefficient of ⫺1. The easiest way to understand this is to plot the returns of two assets against each other for many periods, as is done in Figures 3-3, 3-4, and 3-5.

Most of the points lie on nearly a straight line; a poor return for one was invariably associated with a poor return for the other. The correlation coefficient of .777 for these two assets is quite high. This graph demonstrates that adding U.S. small stocks to a portfolio of U.S. large stocks does not diminish risk very much, as a poor return for one will be very likely associated with a poor return for the other. Figure 3-4 plots two loosely correlated assets—U.S. large stocks (S&P 500) and foreign large stocks (EAFE Index). Although there does appear to be a loose relation between the two, it is far from perfect. The correlation coefficient of this pair is .483. Lastly, Figure 3-5 plots two very poorly correlated assets (correlation coefficient of .068): Japanese small stocks and REITs.

Commodity Trading Advisors: Risk, Performance Analysis, and Selection by Greg N. Gregoriou, Vassilios Karavas, François-Serge Lhabitant, Fabrice Douglas Rouah

Heteroskedasticity was created by letting the values of σ be 5, 10, 15, and 20, with one-fourth of the observations using each value. This allowed us to compare the Spearman correlation coefficient calculated for data sets with and without homoskedasticity. The funds were ranked in ascending order of returns for period one (first 12 months) and period two (last 12 months). From each 24-month period of generated returns, Spearman correlation coefficients were calculated for a fund’s rank in both periods. For the distribution of Spearman correlation coefficients to be suitably approximated by a normal, at least 10 observations are needed. Because 120 pairs are used here, the normal approximation is used.

Table 6.3 examines the correlation coefficients between the different CTA indices as well as between the CTA indices and the first two return moments of the Russell 3000 (Russell squared). The results for the entire sample as well as the subsamples confirm our earlier findings. The correlation coefficient between the CTA index, the Financial and Metal Traders Index, the Systematic Traders Index, and the Diversified Traders Index are positive and close to 1 for all the different periods. The Currency Trader Index and the Discretionary Index have the lowest correlation coefficient with the other CTA indices.

Returns-enhancing diversifiers possess correlations with the same generic asset class in an up market but are relatively less correlated in a down market. 3. “Ineffective” diversifiers are assets that do not add value, even though they may possess significant correlation coefficients with the generic asset class. CTA Strategies for Returns-Enhancing Diversification 339 To illustrate, a hedge fund strategy that has a negative correlation coefficient in an up-market regime and positive correlation coefficient in a down-market regime provides diversification with no incremental returns. We classify this in the third category, that is, as an ineffective diversifier. Indeed, a strategy with such a characteristic will have the opposite effect of a good diversifier as it weakens the returns on an uptrend and exaggerates the negative returns of the portfolio.

pages: 755 words: 121,290

Statistics hacks by Bruce Frey

The Association's marketer has support for her hypothesis. Computing a Correlation Coefficient Just eyeballing two columns of numbers from a sample, though, is usually not enough to really know whether there is a relationship between two things. The marketing specialist in our example wants to use a single number to more precisely describe whatever relationship is seen. The correlation coefficient takes into account all the information we used when we looked at our two columns of numbers in Table 2-1 and decided whether there was a relationship there. The correlation coefficient is produced through a formula that does the following things: Looks at each score in a column Sees how distant that score is from the mean of that column Identifies the distance from the mean of its matching score in the other column Multiplies the paired distances together Averages the results of those multiplications If this were a statistics textbook, I'd have to present a somewhat complicated formula for calculating the correlation coefficient.

This is very close to 1.0, which is the strongest a positive correlation can be, so the cheese-to-cheesecake correlation represents a very strong relationship. Interpreting a Correlation Coefficient Somewhat magically, the correlation formula process produces a number, ranging in value from -1.00 to +1.00, that measures the strength of relationship between two variables. Positive signs indicate the relationship is in the same direction. As one value increases, the other value increases. Negative signs indicate the relationship is in the opposite direction. As one value increases, the other value decreases. An important point to make is that the correlation coefficient provides a standardized measure of the strength of linear relationship between two variables [Hack #12].

Let's imagine that a small college decides to use scores on the American College Test (ACT) as a predictor of college grade point average (GPA) at the end of students' first years. The admissions office goes back through a few years of records and gathers the ACT scores and freshman GPAs for a couple hundred students. They discover, to their delight, that there is a moderate relationship between these two variables: a correlation coefficient of .55. Correlation coefficients are a measure of the strength of linear relationships between two variables [Hack #11], and .55 indicates a fairly large relationship. This is good news because the existence of a relationship between the two makes ACT scores a good candidate as a predictor to guess GPA. Simple linear regression is the procedure that produces all the values we need to cook up the magic formula that will predict the future.

pages: 442 words: 94,734

The Art of Statistics: Learning From Data by David Spiegelhalter

parameters: the unknown quantities in a statistical model, generally denoted with Greek letters. Pearson correlation coefficient: for a set of n paired numbers, (x1, y1), (x2, y2) … (xn, yn), when , sx are the sample mean and standard deviation of the xs, and , sy are the sample mean and standard deviation of the ys, the Pearson correlation coefficient is given by Suppose xs and ys have both been standardized to Z-scores given by us and vs respectively, so that ui = (xi – )/sx, and vi = (yi – )/sy. Then the Pearson correlation coefficient can be expressed as , that is the ‘cross-product’ of the Z-scores. percentile (of a population): there is, for example, a 70% chance of drawing a random observation below the 70th percentile.

Figure 2.6 Two sets of (fictitious) data-points for which the Pearson correlation coefficients are both 0. This clearly does not mean there is no relationship between the two variables being plotted. From Alberto Cairo’s wonderful Datasaurus Dozen4. The Pearson correlation is 0.17 for the 2012–2015 data in Figure 2.5(b), and the Spearman’s rank correlation is −0.03, suggesting that there is no longer any clear relationship between the number of cases and survival rates. However, with so few hospitals the correlation coefficient can be very sensitive to individual data-points – if we remove the smallest hospital, which has a high survival rate, the Pearson correlation jumps to 0.42.

‘Correlation Does Not Imply Causation’ We saw in the last chapter how Pearson’s correlation coefficient measures how close the points on a scatter-plot are to a straight line. When considering English hospitals conducting children’s heart surgery in the 1990s, and plotting the number of cases against their survival, the high correlation showed that bigger hospitals were associated with lower mortality. But we could not conclude that bigger hospitals caused the lower mortality. This cautious attitude has a long pedigree. When Karl Pearson’s newly developed correlation coefficient was being discussed in the journal Nature in 1900, a commentator warned that ‘correlation does not imply causation’.

pages: 404 words: 92,713

The Art of Statistics: How to Learn From Data by David Spiegelhalter

parameters: the unknown quantities in a statistical model, generally denoted with Greek letters. Pearson correlation coefficient: for a set of n paired numbers, (x1, y1), (x2, y2)… (xn, yn), when , sx are the sample mean and standard deviation of the xs, and , sy are the sample mean and standard deviation of the ys, the Pearson correlation coefficient is given by Suppose xs and ys have both been standardized to Z-scores given by us and vs respectively, so that ui = (xi − )/sx, and vi = (yi − )/sy. Then the Pearson correlation coefficient can be expressed as , that is the ‘cross-product’ of the Z-scores. percentile (of a population): there is, for example, a 70% chance of drawing a random observation below the 70th percentile.

Figure 2.6 Two sets of (fictitious) data-points for which the Pearson correlation coefficients are both 0. This clearly does not mean there is no relationship between the two variables being plotted. From Alberto Cairo’s wonderful Datasaurus Dozen4. The Pearson correlation is 0.17 for the 2012–2015 data in Figure 2.5(b), and the Spearman’s rank correlation is −0.03, suggesting that there is no longer any clear relationship between the number of cases and survival rates. However, with so few hospitals the correlation coefficient can be very sensitive to individual data-points—if we remove the smallest hospital, which has a high survival rate, the Pearson correlation jumps to 0.42.

‘Correlation Does Not Imply Causation’ We saw in the last chapter how Pearson’s correlation coefficient measures how close the points on a scatter-plot are to a straight line. When considering English hospitals conducting children’s heart surgery in the 1990s, and plotting the number of cases against their survival, the high correlation showed that bigger hospitals were associated with lower mortality. But we could not conclude that bigger hospitals caused the lower mortality. This cautious attitude has a long pedigree. When Karl Pearson’s newly developed correlation coefficient was being discussed in the journal Nature in 1900, a commentator warned that ‘correlation does not imply causation’.

Stocks for the Long Run, 4th Edition: The Definitive Guide to Financial Market Returns & Long Term Investment Strategies by Jeremy J. Siegel

The diversifying strength of an asset is measured by the correlation coefficient. The correlation coefficient, which theoretically ranges between –1 and +1, measures the correlation between an asset’s return and the return of the rest of the portfolio. The lower the correlation coefficient, the better the asset serves as a portfolio diversifier. Assets with negative correlations are particularly good diversifiers. As the correlation coefficient between the asset and portfolio returns increases, the diversifying quality of the asset declines. The correlation coefficient between annual stock and bond returns for six subperiods between 1926 and 2006 is shown in Figure 2-4.

The correlation of returns between stocks or portfolios of stocks is measured by the correlation coefficient. A good case for investors is if there is no correlation between the stock returns of two countries, and the correlation coefficient is equal to zero. In this case, an investor who allocates his or her portfolio equally between each country can reduce his or her risk by almost one-third, compared to investing in a single country. As the correlation coefficient increases, the gains from diversification dwindle, and if there is perfect synchronization of returns, the correlation coefficient equals 1 and there is no gain (but no loss) from diversification.

For every percentage point increase in the expected risk of U.S. returns or decrease in the expected risk of EAFE returns, the U.S. allocation falls 6.5 percentage points. And a 0.10 increase in the correlation coefficient between EAFE and U.S. returns will lower the EAFE allocation by just over 2 percentage points.8 7 Readers who wish to understand risk-return analysis can go to Richard A. Brealey, Stewart C. Myers, and Franklin Allen, Principles of Corporate Finance,” 8th ed., New York: McGraw-Hill, 2006. 8 The impact of a change in the correlation coefficient is highly nonlinear. If the correlation rises to 0.77, the U.S. allocation will rise over 6 percentage points; if it rises to 0.87, the U.S. share will rise by over 17 percentage points.

pages: 519 words: 102,669

Programming Collective Intelligence by Toby Segaran

Euclidean distance A clear implementation of this formula is shown here: def euclidean(p,q): sumSq=0.0 # add up the squared differences for i in range(len(p)): sumSq+=(p[i]-q[i])**2 # take the square root return (sumSq**0.5) Euclidean distance is used in several places in this book to determine how similar two items are. Pearson Correlation Coefficient The Pearson correlation coefficient is a measure of how highly correlated two variables are. It is a value between 1 and −1, where 1 indicates that the variables are perfectly correlated, 0 indicates no correlation, and −1 means they are perfectly inversely correlated. Figure B-2 shows the Pearson correlation coefficient. Figure B-2. Pearson correlation coefficient This can be implemented with the following code: def pearson(x,y): n=len(x) vals=range(n) # Simple sums sumx=sum([float(x[i]) for i in vals]) sumy=sum([float(y[i]) for i in vals]) # Sum up the squares sumxSq=sum([x[i]**2.0 for i in vals]) sumySq=sum([y[i]**2.0 for i in vals]) # Sum up the products pSum=sum([x[i]*y[i] for i in vals]) # Calculate Pearson score num=pSum-(sumx*sumy/n) den=((sumxSq-pow(sumx,2)/n)*(sumySq-pow(sumy,2)/n))**.5 if den==0: return 1 r=num/den return r We used the Pearson correlation in Chapter 2 to calculate the level of similarity between people's preferences.

, Exercises marketing, Other Uses for Learning Algorithms mass-and-spring algorithm, The Layout Problem matchmaker dataset, Matchmaker Dataset, Difficulties with the Data, Decision Tree Classifier, Categorical Features, Creating the New Dataset, Creating the New Dataset, Applying SVM to the Matchmaker Dataset categorical features, Categorical Features creating new, Creating the New Dataset decision tree algorithm, Decision Tree Classifier difficulties with data, Difficulties with the Data LIBSVM, applying to, Applying SVM to the Matchmaker Dataset scaling data, Creating the New Dataset matchmaker.csv file, Matchmaker Dataset mathematical formulas, Euclidean Distance, Euclidean Distance, Pearson Correlation Coefficient, Weighted Mean, Tanimoto Coefficient, Conditional Probability, Gini Impurity, Entropy, Variance, Gaussian Function, Dot-Products conditional probability, Conditional Probability dot-product, Dot-Products entropy, Entropy Euclidean distance, Euclidean Distance Gaussian function, Gaussian Function Gini impurity, Gini Impurity Pearson correlation coefficient, Pearson Correlation Coefficient Tanimoto coefficient, Tanimoto Coefficient variance, Variance weighted mean, Weighted Mean matplotlib, Graphing the Probabilities, matplotlib, Installation, Simple Usage Example installation, Installation usage example, Simple Usage Example matrix math, Clustering, A Quick Introduction to Matrix Math, A Quick Introduction to Matrix Math, What Does This Have to Do with the Articles Matrix?

pages: 321

Finding Alphas: A Quantitative Approach to Building Trading Strategies by Igor Tulchinsky

For example, the information ratio is just the average returns divided by the standard deviation of returns. Another key quality of an alpha is its uniqueness, which is evaluated by the correlation coefficient between a given alpha and other existing alphas. An alpha with a lower correlation coefficient normally is considered to be adding more value to the pool of existing alphas. If the number of alphas in the pool is small, the importance of correlation is low. As the number of alphas increases, however, different techniques to measure the correlation coefficient among them become more important in helping the investor diversify his or her portfolio. Portfolio managers will want to include relatively uncorrelated alphas in their portfolios because a diversified portfolio helps to reduce risk.

Published 2020 by John Wiley & Sons, Ltd. 62 Finding Alphas where Pit and Pjt denote the PnLs of i th and j th alphas on the t th day, n is the number of days used to measure correlation, and T denotes the matrix transposition. Note: tests usually select the number of days for correlation as two or four years instead of a full history, to save computational resources. Pearson Correlation Coefficient The Pearson correlation coefficient, also known as the Pearson product-­ moment correlation coefficient, has no units and can take values from 1 to 1. The mathematical formula was first developed by Karl Pearson in 1895: cov Pi , Pj r Pi where cov Pi , Pj E Pi Pj Pi (2) Pj Pj is the covariance and Pi and Pj are the standard deviations of Pi and Pj , respectively.

In particular, n r t 1 n t 1 Pit Pit Pi Pi 2 Pjt n t 1 Pj Pjt Pj 2 . (3) The coefficient is invariant to linear transformations of either variable. If the sign of the correlation coefficient is positive, it means that the PnLs of the two alphas tend to move in the same direction. When the return on Pi is positive (negative), the return on Pj has a tendency to be positive (negative) as well. Conversely, a negative correlation coefficient shows that the PnLs of the two alphas tend to move in opposite directions. A zero correlation implies that there is no relationship between two PnL vectors. Figure 8.1 shows the variation of maximum correlation as a function of trading signals, using two years’ worth of data.

pages: 312 words: 35,664

The Mathematics of Banking and Finance by Dennis W. Cox, Michael A. A. Cox

So, from above we know that the gradient of the line is estimated by n n n 1 xi yi − xi yi n i=1 i=1 i=1 â = n n 2 2 1 xi − xi n i=1 i=1 This may be written in a more compact form since the variance (section 5.6) is n n 2 2 1 xi − xi n i=1 i=1 var(x) = n−1 The covariance may be similarly written as: n cov(x, y) = (xi − x̄) (yi − ȳ) i=1 = n−1 n xi yi i=1 n n 1 − xi yi n i=1 i=1 n−1 (*) We can then write the equation for the estimated gradient in a more compact form using this new notation: â = cov(x, y) var(x) 13.3 CORRELATION COEFFICIENT A closely related term to the covariance is the correlation coefficient (r (x, y)), which is simply r (x, y) = cov(x, y) std(x) std(y) where the standard deviation of x (section 5.6) is given by n 2 n 2 1 i=1 xi − n i=1 xi std(x) = n−1 There would be a similar expression for y. The correlation coefficient is a measure of the interdependence of two variables. The coefficient ranges in value from −1 to +1, indicating perfect negative correlation at −1, absence Linear Regression 105 of correlation at zero, and perfect positive correlation at +1.

The coefficient ranges in value from −1 to +1, indicating perfect negative correlation at −1, absence Linear Regression 105 of correlation at zero, and perfect positive correlation at +1. Two variables are positively correlated if the correlation coefficient is greater than zero and the line that is drawn to show a relationship between the items sampled has a positive gradient. If the correlation coefficient is negative, the variables are negatively correlated and the line drawn will have a negative gradient. The final option is that the correlation coefficient vanishes, and the gradient vanishes since â = 0, in which case the variables are completely uncorrelated. This means that there is no relationship between the two variables.

pages: 447 words: 104,258

Mathematics of the Financial Markets: Financial Instruments and Derivatives Modelling, Valuation and Risk Issues by Alain Ruttiens

The linear correlation of x and y is 0, although they are clearly dependent: if x = 0, y can only value 1 or −1, and if x ≠ 0, y = 0. Coming back to the general case of φx(x) and φy(y) being not Gaussian, this inference cannot be made. Typically, the classic “rank correlation” coefficient of Spearman shows the way to get round the problem: this rank correlation consists in a linear correlation coefficient of the variates, 5 now transformed in a non-linear way, by a probability transformation, that is, their respective cumulative marginal distributions: with The Spearman correlation is a correlation measure that can be computed from these relationships and from the general formula for ρx, y above, but, as a step further, we can link above Φx(x), Φy(y) and Φ(x, y) relationships in a more general way that defines C – named a copula of two variables x and y – as a cumulative probability function of the marginal cumulative probabilities Φx(x), Φy(y) of x and y.6 A copula is thus a general measure of co-dependence between two variates, which is independent of their individual marginal distribution – see Figure 13.5.

It is indeed based on several restrictive hypotheses: Hypotheses related to financial assets: Asset returns r are modeled by a random variable, distributed as a Gaussian probabilities distribution, fully determined by its first two moments, namely its expected value E and its variance V, although instead of V, the theory makes use of the corresponding standard deviation measure STD (STD = ). Returns of different financial assets i and j are correlated by the linear correlation coefficient ρij. Markets are efficient1 – practically speaking, we observe that the more liquid a market, the more efficient it is. The theory is built on mid prices (average of the market quoted bid and offer (or ask) prices): the market bid–offer spread is thus not considered here. Various costs such as brokerage fees, taxes, and so on are not taken into account (they are too much affected by local circumstances, market features, and the investor's situation).

For example, in 2006, based on successive daily close prices, the return and risk of L'Oreal were 20% and 19% respectively. Figure 4.3 Example of a stock showed in a (r, σ) graph 4.3.3 The Markowitz model Markowitz's goal was to optimize the budget allocation to a portfolio P of n stocks Si(ri, σi), weighted by wi, with 0 ≤ wi ≤ 1 and ∑wi = 1, so that for P: (4.1) that is, where the ρij correlation coefficients are computed by In a (r, σ) chart, it is possible, for a given past period of data to locate by a point any Si(ri, σi), but also any possible weighted combination of up to n stocks, defining points that represent portfolios, among which the optimal ones have to be identified. Performing this graph representation shows that there is a (non-linear) “frontier” of possible portfolios presenting the highest return, for different risks.

pages: 297 words: 91,141

Market Sense and Nonsense by Jack D. Schwager

Correlation Defined The correlation coefficient, typically denoted by the letter r, measures the degree of linear relationship between two variables. The correlation coefficient ranges from −1.0 to +1.0. The closer the correlation coefficient is to +1.0, the closer the relationship is between the two variables. A perfect correlation of 1.0 would occur only in artificial situations. For example, the heights of a group of people measured in inches and the heights of the same group of people measured in feet would be perfectly correlated. The closer the correlation coefficient is to −1.0, the stronger the inverse correlation is between the two variables.

The closer the correlation coefficient is to −1.0, the stronger the inverse correlation is between the two variables. For example, average winter temperatures in the U.S. Northeast and heating oil usage in that region would be inversely related variables (variables with a negative correlation coefficient). If two variables have a correlation coefficient near zero, it indicates that there is no significant (linear) relationship between the variables. It is important to understand that the correlation coefficient only indicates the degree of correlation between two variables and does not imply anything about cause and effect. Correlation Shows Linear Relationships Correlation reflects only linear relationships.

Because correlation reflects only linear relationships, and there is no linear relationship between the two variables. Figure 9.1 Strategy Returns versus S&P Returns The Coefficient of Determination (r2) The square of the correlation coefficient, which is called the coefficient of determination and is denoted as r2, has a very specific interpretation: It represents the percentage of the variability of one variable explained by the other. For example, if the correlation coefficient (r) of a fund versus the S&P is 0.7, it implies that nearly half the variability of the fund’s returns is explained by the S&P returns (r2 = 0.49). For a mutual fund that is a so-called closet benchmarker—a fund that maintains a portfolio very similar to the S&P index with only minor differences—the r2 would tend to be very high (e.g., above 0.9).

pages: 321 words: 97,661

How to Read a Paper: The Basics of Evidence-Based Medicine by Trisha Greenhalgh

Ignore all withdrawals (‘drop outs’) and non-responders, so the analysis only concerns subjects who fully complied with treatment (see section ‘Were preliminary statistical questions addressed?’). 5. Always assume that you can plot one set of data against another and calculate an ‘r-value’ (Pearson correlation coefficient) (see section ‘Has correlation been distinguished from regression, and has the correlation coefficient (‘r-value’) been calculated and interpreted correctly?’), and that a ‘significant’ r-value proves causation (see section ‘Have assumptions been made about the nature and direction of causality?’). 6. If outliers (points that lie a long way from the others on your graph) are messing up your calculations, just rub them out.

Every r-value should be accompanied by a p-value, which expresses how likely an association of this strength would be to have arisen by chance (see section ‘Have ‘p-values’ been calculated and interpreted appropriately?’), or a confidence interval, which expresses the range within which the ‘true’ R-value is likely to lie (see section ‘Have confidence intervals been calculated, and do the authors' conclusions reflect them?’). (Note that lower case ‘r’ represents the correlation coefficient of the sample, whereas upper case ‘R’ represents the correlation coefficient of the entire population.) Remember, too, that even if the r-value is an appropriate value to calculate from a set of data, it does not tell you whether the relationship, however strong, is causal (see subsequent text). The term regression refers to a mathematical equation that allows one variable (the target variable) to be predicted from another (the independent variable).

As you might imagine, statistical significance is more difficult to demonstrate with rank order tests (indeed, some statisticians are cynical about the value of the latter), and this tempts researchers to use statistics such as the r-value (see section ‘Has correlation been distinguished from regression, and has the correlation coefficient (“r-value”) been calculated and interpreted correctly?’) inappropriately. Not only is the r-value (parametric) easier to calculate than an equivalent rank order statistic such as Spearman's ρ (pronounced ‘rho’) but it is also much more likely to give (apparently) significant results. Unfortunately, it will also give an entirely spurious and misleading estimate of the significance of the result, unless the data are appropriate to the test being used.

pages: 119 words: 10,356

Topics in Market Microstructure by Ilija I. Zovko

Later in the text we use a bootstrap approach to test the significance. Now, however, we test the significance of the correlation coefficients using a standard algorithm as in ref. (Best and Roberts, 1975). The algorithm calculates the approximate tail probabilities for Spearman’s correlation coefficient ρ. Its precision unfortunately degrades when there are ties in the data, which is the case here. With this caveat in mind, as a preliminary test, we find that, for example, for on-book trading in Vodafone for the month of May 2000, 10.3% of all correlation coefficients are significant at the 5% level. Averaging over all stocks and months, the average percentage of significant coefficients for on-book trading is 10.5% ± 0.4%, while for off-book trading it is 20.7% ± 1.7%.

Averaging over all stocks and months, the average percentage of significant coefficients for on-book trading is 10.5% ± 0.4%, while for off-book trading it is 20.7% ± 1.7%. Both of these averages are substantially higher than the 5% we would expect randomly with a 5% acceptance level of the test. 4.2 Significance and structure in the correlation matrices The preliminary result of the previous section that some correlation coefficients are non-random is further corroborated by testing for non-random structure in the correlation matrices. The hypothesis that there is structure in the correlation matrices contains the weaker hypothesis that some coefficients are statistically significant. The test for structure in the matrices would involve multiple joint tests for the significance of the coefficients.

We apply clustering techniques using a metric chosen so that two strongly correlated institutions are ’close’ and anti-correlated institutions are ’far away’. A functional form fulfilling this requirement and satisfying the properties of being a metric is (Bonanno et al., 2000) # (4.2) di,j = 2 · (1 − ρi,j ), where ρi,j is the correlation coefficient between strategies i and j. We have tried several reasonable modifications to this form but without obvious differences in the results. Ultimately the choice of this metric is influenced by the fact that it has been successfully used in other studies (Bonanno et al., 2000). We use complete linkage clustering, in which the distance between two clusters is calculated as the maximum distance between its members.

Hands-On Machine Learning With Scikit-Learn, Keras, and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems by Aurelien Geron

Finally, coefficients close to zero mean that there is no linear correlation. Figure 2-14 shows various plots along with the correlation coefficient between their horizontal and vertical axes. Figure 2-14. Standard correlation coefficient of various datasets (source: Wikipedia; public domain image) Warning The correlation coefficient only measures linear correlations (“if x goes up, then y generally goes up/down”). It may completely miss out on nonlinear relationships (e.g., “if x is close to zero then y generally goes up”). Note how all the plots of the bottom row have a correlation coefficient equal to zero despite the fact that their axes are clearly not independent: these are examples of nonlinear relationships.

Looking for Correlations Since the dataset is not too large, you can easily compute the standard correlation coefficient (also called Pearson’s r) between every pair of attributes using the corr() method: corr_matrix = housing.corr() Now let’s look at how much each attribute correlates with the median house value: >>> corr_matrix["median_house_value"].sort_values(ascending=False) median_house_value 1.000000 median_income 0.687170 total_rooms 0.135231 housing_median_age 0.114220 households 0.064702 total_bedrooms 0.047865 population -0.026699 longitude -0.047279 latitude -0.142826 Name: median_house_value, dtype: float64 The correlation coefficient ranges from –1 to 1.

Note how all the plots of the bottom row have a correlation coefficient equal to zero despite the fact that their axes are clearly not independent: these are examples of nonlinear relationships. Also, the second row shows examples where the correlation coefficient is equal to 1 or –1; notice that this has nothing to do with the slope. For example, your height in inches has a correlation coefficient of 1 with your height in feet or in nanometers. Another way to check for correlation between attributes is to use Pandas’ scatter_matrix function, which plots every numerical attribute against every other numerical attribute. Since there are now 11 numerical attributes, you would get 112 = 121 plots, which would not fit on a page, so let’s just focus on a few promising attributes that seem most correlated with the median housing value (Figure 2-15): from pandas.plotting import scatter_matrix attributes = ["median_house_value", "median_income", "total_rooms", "housing_median_age"] scatter_matrix(housing[attributes], figsize=(12, 8)) Figure 2-15.

Calling Bullshit: The Art of Scepticism in a Data-Driven World by Jevin D. West, Carl T. Bergstrom

Quarterbacks and kickers, for example, tend to be lighter than you would expect given their height, whereas running backs and linemen tend to be heavier. The strength of a linear correlation is measured as a correlation coefficient, which is a number between 1 and −1. A correlation of 1 means that the two measurements form a perfect line on a scatter plot, such that when one increases, the other increases as well. For example, distance in meters and distance in kilometers have a correlation coefficient of 1, because the former is just one thousand times the latter. A correlation of −1 means that two measurements form another kind of perfect line on a scatter plot, such that when one increases, the other decreases.

As one increases, the other decreases by the same amount. These two quantities have a correlation of −1. A correlation coefficient of 0 means that a best-fit line through the points doesn’t tell you anything.*2 In other words, one measurement tells you nothing about the other.*3 For example, psychologists sometimes use the Eysenck Personality Inventory questionnaire as a way to summarize aspects of personality known as impulsivity, sociability, and neuroticism. Across individuals, impulsivity and neuroticism are essentially uncorrelated, with a correlation coefficient of −0.07. In other words, knowing something about a person’s impulsivity tells you very little (if anything) about his neuroticism and vice versa.

For example, consider pairs of numbers {x, sin(x)}. If we know x, we can predict exactly what sin(x) will be, but the correlation coefficient—a measure of linear correlation—between these numbers is zero across a full-cycle sine wave. There is no linear correlation between x and sin(x) because a best-fit line through {x, sin(x)} pairs has a slope of 0 and tells us nothing about the likely value of sin(x) for any given value of x. *2 In the rare case where the data points form either a vertical or horizontal line, the correlation coefficient is undefined. In these cases, knowing one measurement also tells you nothing about the other measurement

pages: 436 words: 140,256

The Rise and Fall of the Third Chimpanzee by Jared Diamond

A simple numerical way of describing the result is by means of a statistical index called the correlation coefficient. If you line up 100 husbands in order of their ranking for some characteristic (say, their height), and if you also line up their 100 wives with respect to the same characteristic, the correlation coefficient describes whether a man tends to be at the same position in the husbands' line-up as his wife is in the line-up of wives. A correlation coefficient of plus one would mean perfect correspondence: the tallest man marries the tallest woman, the thirty-seventh tallest man marries the thirty-seventh tallest woman, and so on. A correlation coefficient of minus one would mean perfect matching by opposites: the tallest man marries the shortest woman, the thirty-seventh tallest man marries the thirty-seventh shortest woman, and so on.

A correlation coefficient of minus one would mean perfect matching by opposites: the tallest man marries the shortest woman, the thirty-seventh tallest man marries the thirty-seventh shortest woman, and so on. Finally, a correlation coefficient of zero would mean that husbands and wives assort completely randomly by height: a tall man is as likely to marry a short woman as a tall woman. These examples are for height, but correlation coefficients can also be calculated for anything else, such as income and IQ. If you measure enough things about enough couples, here is what you will find. Not surprisingly, the highest correlation coefficients—typically around +0.9—are for religion, ethnic background, race, socioeconomic status, age, and political views.

Not surprisingly, the highest correlation coefficients—typically around +0.9—are for religion, ethnic background, race, socioeconomic status, age, and political views. That is, most husbands and wives prove to be of the same religion, ethnic background, and so on. Perhaps you also will not be surprised that the next highest correlation coefficients, usually around +0.4, are for measures of personality and intelligence, such as extroversion, neatness, and IQ. Slobs tend to marry slobs, though the chances of a slob marrying a compulsively neat person are not as low as the chances of a political reactionary marrying a left-winger. What about matching of husbands and wives for physical characteristics? The answer is not one that would leap out at you immediately if you just looked at a few married couples.

pages: 517 words: 139,477

Stocks for the Long Run 5/E: the Definitive Guide to Financial Market Returns & Long-Term Investment Strategies by Jeremy Siegel

This will be particularly true if bond and stock returns are negatively correlated, which would happen if bond and stock prices move in the opposite direction.4 The diversifying strength of an asset is measured by the correlation coefficient. The correlation coefficient ranges between -1 and +1 and measures the co-movement between an asset’s return and the return of the rest of the portfolio. The lower the correlation coefficient, the better the asset serves as a portfolio diversifier. Assets with near-zero or especially negative correlations are particularly good diversifiers. As the correlation coefficient between the asset and portfolio returns increases, the diversifying quality of the asset declines. In Chapter 3 we examined the changing correlation coefficient between the return on 10-year Treasury bonds and stocks, represented by the S&P 500 Index.

The standard deviation of the Magellan Fund over Lynch’s period was 21.38 percent, compared with 13.88 percent for the Wilshire 5000, while its correlation coefficient with the Wilshire was .86. 6. “The Superinvestors of Graham-and-Doddsville,” Hermes, the Columbia Business School Magazine, 1984 (reprinted 2004). 7. Money managers are assumed to expose their clients to the same risk as would the market, and the money managers have a correlation coefficient of .88 with market returns, which has been typical of equity mutual funds since 1971. 8. Darryll Hendricks, Jayendu Patel, and Richard Zeckhauser, “Hot Hands in Mutual Funds: Short-Run Persistence of Relative Performance, 1974-1988,” Journal of Finance, vol. 48, no. 1 (March 1993), pp. 93-130. 9.

In Chapter 3 we examined the changing correlation coefficient between the return on 10-year Treasury bonds and stocks, represented by the S&P 500 Index. Figure 6-3 displays the correlation coefficient between annual stock and bond returns for three subperiods between 1926 and 2012. From 1926 through 1965 the correlation was only slightly positive, indicating that bonds were fairly good diversifiers for stocks. Bonds were good diversifiers in this period because it contained the Great Depression, which was characterized by falling economic activity and consumer prices, a situation that was bad for stocks but good for U.S. government bonds. FIGURE 6-3 Correlation of Real Bond and Stock Returns Over Various Historical Periods However, under a paper money standard, bad economic times are more likely to be associated with inflation, not deflation.

pages: 239 words: 77,436

Pure, White and Deadly: How Sugar Is Killing Us and What We Can Do to Stop It by John Yudkin

If this were so, you would say that the correlation coefficient was 1·0. Supposing on the other hand – and this is even more unlikely – that there was no relationship whatever between height and weight, so that it would be just as likely for a man weighing 150 pounds to be five feet tall or six feet tall. In this case the correlation coefficient would be 0. In fact, there is a relationship, but not a precise one; tall people tend to be heavier. If you work it out exactly, for adult men the correlation coefficient between height and weight comes to about 0·6. The correlation coefficients I have found so far for cancer and sugar consumption in different countries are as follows: Cancer of the large intestine in men: 0·60 Cancer of the large intestine in women: 0·50 Cancer of the breast: 0·63 However, such international statistics, as I have stressed repeatedly, can do no more than give a clue as to the possible role of sugar or fat in producing disease.

On the other hand, the lowest mortality, in ascending order, is in Japan, Yugoslavia, Portugal, Spain and Italy, with the lowest sugar consumption in Japan, Portugal, Spain, Yugoslavia and Italy. My own observations of the association between sugar and cancer of the intestine and of the breast were made several years earlier than the study I have just quoted. I calculated what are called the ‘correlation coefficients’ between these cancers and sugar consumption in all the countries for which statistics were then available. Let me explain first what correlation coefficients are, and let me take as an example the relation between people’s height and weight. On the whole, the taller people are, the more they weigh. But it is all very well to say that there is ‘on the whole’ this association between height and weight; it would be better if we could say how close this association is.

pages: 589 words: 69,193

Mastering Pandas by Femi Anthony

The correlation measure, known as correlation coefficient, is a number that captures the size and direction of the relationship between the two variables. It can vary from -1 to +1 in direction and 0 to 1 in magnitude. The direction of the relationship is expressed via the sign, with a + sign expressing positive correlation and a - sign negative correlation. The higher the magnitude, the greater the correlation with a one being termed as the perfect correlation. The most popular and widely used correlation coefficient is the Pearson product-moment correlation coefficient, known as r. It measures the linear correlation or dependence between two x and y variables and takes values between -1 and +1.

We can also only make predictions for values within the bounds of the data. For example, we cannot predict what the chirpFrequency is at 32 degrees Fahrenheit as it is outside the bounds of the data; moreover, at 32 degrees Fahrenheit, the crickets would have frozen to death. The value of R, the correlation coefficient, is given as follows: In [38]: R=np.sqrt(result.rsquared) R Out[38]: 0.83514378678237422 Thus, our correlation coefficient is R = 0.835. This would indicate that about 84 percent of the chirp frequency can be explained by the changes in temperature. Reference of this information: The Song of Insects http://www.hup.harvard.edu/catalog.php?isbn=9780674420663 The data is sourced from http://bit.ly/1MrlJqR.

It measures the linear correlation or dependence between two x and y variables and takes values between -1 and +1. The sample correlation coefficient r is defined as follows: This can also be written as follows: Here, we have omitted the summation limits. Linear regression As mentioned earlier, regression focuses on using the relationship between two variables for prediction. In order to make predictions using linear regression, the best-fitting straight line must be computed. If all the points (values for the variables) lie on a straight line, then the relationship is deemed perfect. This rarely happens in practice and the points do not all fit neatly on a straight line.

pages: 147 words: 39,910

The Great Mental Models: General Thinking Concepts by Shane Parrish

The problem is, without a good understanding of what is meant by these terms, these decisions fail to capitalize on real dynamics in the world and instead are successful only by luck. No Correlation The correlation coefficient between two measures, which varies between -1 and 1, is a measure of the relative weight of the factors they share. For example, two phenomena with few factors shared, such as bottled water consumption versus suicide rate, should have a correlation coefficient of close to 0. That is to say, if we looked at all countries in the world and plotted suicide rates of a specific year against per capita consumption of bottled water, the plot would show no pattern at all.

The only factor governing temperature—velocity of molecules—is shared by all scales. Thus each degree in Celsius will have exactly one corresponding value in Fahrenheit. Therefore temperature in Celsius and Fahrenheit will have a correlation coefficient of 1 and the plot will be a straight line. Weak to Moderate Correlation There are few phenomena in human sciences that have a correlation coefficient of 1. There are, however, plenty where the association is weak to moderate and there is some explanatory power between the two phenomena. Consider the correlation between height and weight, which would land somewhere between 0 and 1.

Risk Management in Trading by Davis Edwards

(See Figure 3.10, Positive and Negative Correlation.) Some features of correlation are: ■ ■ ■ Positive Correlation. A correlation coefficient equal to +1 means that the two series have behaved identically over the testing period. Negative Correlation. A correlation coefficient of −1 indicates that the series have been inversely proportional during the testing period. In other words, when one price rises, the other price falls. Zero Correlation. A correlation coefficient of zero indicates no relationship between the two values The calculation of the correlation coefficient, ρ, is mathematically defined. (See Equation 3.10, Correlation.) Positive Correlation FIGURE 3.10 Negative Correlation Positive and Negative Correlation ∑ ρ= (x−x) (y − y ) (n − 1)σ x σ y where x Data Set.

If crude oil prices rise, then the stock price of the oil company should also be likely to rise. When determining if two prices are correlated, it is necessary to compare changes in price rather than prices. The most common way to measure the relationship between two assets is to calculate the correlation coefficient of their price changes. The correlation coefficient is a number between −1 and +1 that indicates the strength of 77 Financial Mathematics KEY CONCEPT: CORRELATION In the financial markets, the statement that “two assets are correlated” means “the price changes in the two assets are correlated” rather than the “prices are correlated.”

The relative size of the value assigned to the hedge and hedged item do not affect hedge effectiveness. A number of summary statistics are produced by a regression analysis. These statistics are commonly used to evaluate the effectiveness of the hedge. Two major statistics used for this purpose are the slope (abbreviated b above) and the correlation coefficient (usually squared and abbreviated as R2 or R‐squared). Secondary statistics include checking that there are enough observations to conduct a valid test, and that the slope and R2 tests are sufficiently stable to trust the results. For example, a hedge‐accounting memo might define five tests to determine a highly effective hedge.

pages: 482 words: 121,672

A Random Walk Down Wall Street: The Time-Tested Strategy for Successful Investing (Eleventh Edition) by Burton G. Malkiel

Some portfolio managers have argued that diversification has not continued to give the same degree of benefit as was previously the case. Globalization led to an increase in the correlation coefficients between the U.S. and foreign markets as well as between stocks and commodities. The following three charts indicate how correlation coefficients have risen over the first decade of the 2000s. The charts show the correlation coefficients calculated over every twenty-four-month period between U.S. stocks (as measured by the S&P 500-Stock Index) and the EAFE index of developed foreign stocks, between U.S. stocks and the broad (MSCI) index of emerging-market stocks, and between U.S. stocks and the Goldman Sachs (GSCI) index of a basket of commodities such as oil, metals, and the like.

The example may seem a bit strained, and most investors will realize that when the market gets clobbered, just about all stocks go down. Still, at least at certain times, some stocks and some classes of assets do move against the market; that is, they have negative covariance or (and this is the same thing) they are negatively correlated with each other. THE CORRELATION COEFFICIENT AND THE ABILITY OF DIVERSIFICATION TO REDUCE RISK Correlation Coefficient Effect of Diversification on Risk +1.0 No risk reduction is possible. +0.5 Moderate risk reduction is possible. 0 Considerable risk reduction is possible. –0.5 Most risk can be eliminated. –1.0 All risk can be eliminated. Now comes the real kicker; negative correlation is not necessary to achieve the risk reduction benefits from diversification.

Now comes the real kicker; negative correlation is not necessary to achieve the risk reduction benefits from diversification. Markowitz’s great contribution to investors’ wallets was his demonstration that anything less than perfect positive correlation can potentially reduce risk. His research led to the results presented in the preceding table. As shown, it demonstrates the crucial role of the correlation coefficient in determining whether adding a security or an asset class can reduce risk. DIVERSIFICATION IN PRACTICE To paraphrase Shakespeare, can there be too much of a good thing? In other words, is there a point at which diversification is no longer a magic wand safeguarding returns? Numerous studies have demonstrated that the answer is yes.

pages: 1,331 words: 163,200

Hands-On Machine Learning With Scikit-Learn and TensorFlow: Concepts, Tools, and Techniques to Build Intelligent Systems by Aurélien Géron

Finally, coefficients close to zero mean that there is no linear correlation. Figure 2-14 shows various plots along with the correlation coefficient between their horizontal and vertical axes. Figure 2-14. Standard correlation coefficient of various datasets (source: Wikipedia; public domain image) Warning The correlation coefficient only measures linear correlations (“if x goes up, then y generally goes up/down”). It may completely miss out on nonlinear relationships (e.g., “if x is close to zero then y generally goes up”). Note how all the plots of the bottom row have a correlation coefficient equal to zero despite the fact that their axes are clearly not independent: these are examples of nonlinear relationships.

Looking for Correlations Since the dataset is not too large, you can easily compute the standard correlation coefficient (also called Pearson’s r) between every pair of attributes using the corr() method: corr_matrix = housing.corr() Now let’s look at how much each attribute correlates with the median house value: >>> corr_matrix["median_house_value"].sort_values(ascending=False) median_house_value 1.000000 median_income 0.687170 total_rooms 0.135231 housing_median_age 0.114220 households 0.064702 total_bedrooms 0.047865 population -0.026699 longitude -0.047279 latitude -0.142826 Name: median_house_value, dtype: float64 The correlation coefficient ranges from –1 to 1.

Note how all the plots of the bottom row have a correlation coefficient equal to zero despite the fact that their axes are clearly not independent: these are examples of nonlinear relationships. Also, the second row shows examples where the correlation coefficient is equal to 1 or –1; notice that this has nothing to do with the slope. For example, your height in inches has a correlation coefficient of 1 with your height in feet or in nanometers. Another way to check for correlation between attributes is to use Pandas’ scatter_matrix function, which plots every numerical attribute against every other numerical attribute. Since there are now 11 numerical attributes, you would get 112 = 121 plots, which would not fit on a page, so let’s just focus on a few promising attributes that seem most correlated with the median housing value (Figure 2-15): from pandas.tools.plotting import scatter_matrix attributes = ["median_house_value", "median_income", "total_rooms", "housing_median_age"] scatter_matrix(housing[attributes], figsize=(12, 8)) Figure 2-15.

pages: 741 words: 199,502

Human Diversity: The Biology of Gender, Race, and Class by Charles Murray

Take a pencil (literally or imaginarily) and draw a straight, sloping line through the dots in a way that seems to you to best reflect this upward-sloping trend. Now continue to read and see how well you have intuitively produced the basis for a correlation coefficient and a regression coefficient. The Correlation Coefficient Modern statistics provide more than one method for measuring correlation, but we confine ourselves to the one that is most important in both use and generality: the Pearson product-moment correlation coefficient (named after Karl Pearson, the English mathematician and biometrician). To get at this coefficient, let us first replot the graph of the class, replacing inches and pounds with standard scores.

We focus on the slope of the best-fitting line because it is the correlation coefficient—in this case, equal to .50, which is quite large by the standards of variables used by social scientists. The closer it gets to ±1.0, the stronger is the linear relationship between the standardized variables (the variables expressed as standard scores). When the two variables are mutually independent, the best-fitting line is horizontal; hence its slope is 0. Anything other than 0 signifies a relationship, albeit possibly a very weak one. Whatever the correlation coefficient of a pair of variables is, squaring it yields another notable number.

“Multiply the effect identified with this correlation by the number of people in a department store the week before Christmas,” the authors wrote, “and it becomes obvious why merchandisers should care deeply about the personalities of their customers.”9 They offered a new set of guidelines based on the correlation coefficient (r). In the summary that follows, I have replaced the value of r with the equivalent value of Cohen’s d. The authors argued that an effect size of .10 “is ‘very small’ for the explanations of single events but potentially consequential in the not-very long run,” while an effect size of .20 “is still ‘small’ at the level of single events but potentially more ultimately consequential.”10 Other scholars have advocated similar guidelines for interpreting small values of d.11 But their treatment of “small” collides with the position taken by the most influential work arguing for small sex differences in cognitive repertoires—the “gender similarities hypothesis” originated by psychologist Janet Shibley Hyde in the September 1985 issue of American Psychologist, the flagship journal of the American Psychological Association.

pages: 408 words: 85,118

Python for Finance by Yuxing Yan

What is the market risk (beta) for IBM in 2010? (Hint: the source of data could be from Yahoo! Finance.) 19. What is wrong with the following lines of code? >>>c=20 >>>npv=np.npv(0.1,c) [ 121 ] Introduction to NumPy and SciPy 20. The correlation coefficient function from NumPy is np.corrcoef(). Find more about this function. Estimate the correlation coefficient between IBM, DELL, and W-Mart. 21. Why is it claimed that the sn.npv() function from SciPY() is really a Present Value (PV) function? 22. Design a true NPV function using all cash flows, including today's cash flow. 23. The Sharpe ratio is used to measure the trade-off between risk and return: Sharpe = R − Rf σ Here, R is the expected returns for an individual security, and R f is the expected risk-free rate. σ is the volatility, that is, standard deviation of the return on the underlying security.

For more detail about the function, just type help(plt.annotate) after issuing import matplotlib.pyplot as plt. From the preceding graph, we see that the fluctuation, uncertainty, or risk of our equal-weighted portfolio is much smaller than those of individual stocks in its portfolio. We can also estimate their means, standard deviation, and correlation coefficient. The correlation coefficient between those two stocks is -0.75, and this is the reason why we could diversify away firm-specific risk by forming an even equal-weighted portfolio as shown in the following code: >>>import scipy as sp >>>sp.corrcoef(A,B) array([[ 1. , -0.74583429], [-0.74583429, 1. ]]) In the preceding example, we use hypothetical numbers (returns) for two stocks.

First, let us look at a hypothetical case by assuming that we have 5 years' annual returns of two stocks as follows: Year Stock A Stock B 2009 0.102 0.1062 2010 -0.02 0.23 2011 0.213 0.045 2012 0.12 0.234 2013 0.13 0.113 We form an equal-weighted portfolio using those two stocks. Using the mean() and std() functions contained in NumPy, we can estimate their means, standard deviations, and correlation coefficients as follows: >>>import numpy as np >>>A=[0.102,-0.02, 0.213,0.12,0.13] >>>B=[0.1062,0.23, 0.045,0.234,0.113] >>>port_EW=(np.array(ret_A)+np.array(ret_B))/2. >>>round(np.mean(A),3),round(np.mean(B),3),round(np.mean(port_EW),3) (0.109, 0.146, 0.127) >>>round(np.std(A),3),round(np.std(B),3),round(np.std(port_EW),3) (0.075, 0.074, 0.027) In the preceding code, we estimate mean returns, their standard deviations for individual stocks, and an equal-weighted portfolio.

pages: 416 words: 118,592

A Random Walk Down Wall Street: The Time-Tested Strategy for Successful Investing by Burton G. Malkiel

Some portfolio managers have argued that diversification has not continued to give the same degree of benefit as was previously the case. Globalization led to an increase in the correlation coefficients between the U.S. and foreign markets as well as between stocks and commodities. The following three charts indicate how correlation coefficients have risen over the first decade of the 2000s. The charts show the correlation coefficients calculated over every twenty-four-month period between U.S. stocks (as measured by the S&P 500-Stock Index) and the EAFE index of developed foreign stocks, between U.S. stocks and the broad (MSCI) index of emerging-market stocks, and between U.S. stocks and the Goldman Sachs (GSCI) index of a basket of commodities such as oil, metals, and the like.

The example may seem a bit strained, and most investors will realize that when the market gets clobbered, just about all stocks go down. Still, at least at certain times, some stocks and some classes of assets do move against the market; that is, they have negative covariance or (and this is the same thing) they are negatively correlated with each other. THE CORRELATION COEFFICIENT AND THE ABILITY OF DIVERSIFICATION TO REDUCE RISK Correlation Coefficient Effect of Diversification on Risk +1.0 No risk reduction is possible. +0.5 Moderate risk reduction is possible. 0 Considerable risk reduction is possible. –0.5 Most risk can be eliminated. –1.0 All risk can be eliminated. Now comes the real kicker; negative correlation is not necessary to achieve the risk reduction benefits from diversification.

Now comes the real kicker; negative correlation is not necessary to achieve the risk reduction benefits from diversification. Markowitz’s great contribution to investors’ wallets was his demonstration that anything less than perfect positive correlation can potentially reduce risk. His research led to the results presented in the preceding table. As shown, it demonstrates the crucial role of the correlation coefficient in determining whether adding a security or an asset class can reduce risk. DIVERSIFICATION IN PRACTICE To paraphrase Shakespeare, can there be too much of a good thing? In other words, is there a point at which diversification is no longer a magic wand safeguarding returns? Numerous studies have demonstrated that the answer is yes.

pages: 369 words: 128,349

Beyond the Random Walk: A Guide to Stock Market Anomalies and Low Risk Investing by Vijay Singal

Notice that the returns from beachwear and video rental always go in the opposite direction. If one of them does well, the other does not. Therefore, adding stocks that do not behave like other stocks in your portfolio is good and can reduce risk. The correlation is measured by what is called a correlation coefficient. The correlation coefficient varies between –1 and +1. The two stocks in the above example have a correlation of –1. Unfortunately, most stocks have a positive correlation, and many of them have a correlation with the market portfolio that is close to +1. The challenge in diversifying risk is to find stocks that have a correlation of less than +1.

Therefore, it is not always necessary to ensure that the return is preserved. A general rule to evaluate whether a new asset should be included in an existing portfolio is based on the risk-return trade-off relationship: E(Rn ) = R f + σ n ρn, p σp × E(Rp ) − R f  where E(R) is the return from an asset, s is the standard deviation, r is the correlation coefficient, and the subscripts n and p refer to the new stock and existing portfolio. Rf is the return on the risk-free asset. If the new asset’s return is greater than the right-hand side in the above equation, then the asset should be included in the existing portfolio, otherwise not. That condition can be rewritten as below: E( Rn ) − R f σn > E(Rp ) − R f σp × ρn, p Evidence Before looking at the evidence, consider the potential benefits from international investing and the source of those benefits.

That condition can be rewritten as below: E( Rn ) − R f σn > E(Rp ) − R f σp × ρn, p Evidence Before looking at the evidence, consider the potential benefits from international investing and the source of those benefits. Assume that the dollar return on U.S. stocks is 12 percent with a standard deviation of 18 percent, and the dollar return on non-U.S. stocks is also 12 percent with a standard deviation of 18 percent. Since the U.S. markets and foreign markets are not well correlated, let the correlation coefficient be 0.60. Putting the U.S. stocks and the non-U.S. stocks in a 50-50 combination would generate a new world portfolio with the following characteristics: Rw = w1RUS + w2 Rnon −US = 0.50 × 12% + 0.50 × 12% = 12% σ w = w12σ 12 + w22σ 22 + 2w1w2 ρσ 1σ 2 = 0.50 2 × 0.18 2 + 0.50 2 × 0.18 2 + 2 × 0.50 × 0.50 × 0.60 × 0.18 × 0.18 = 0.16 236 Beyond the Random Walk The new world portfolio has a return of 12 percent and a risk of 16 percent.

Trading Risk: Enhanced Profitability Through Risk Control by Kenneth L. Grant

For example, 170 TRADING RISK there’s no reason to believe that there are any statistical commonalities between, say, the Swedish rate of inflation and the price of silkworms in Malaysia; and over time we would expect a correlation between these two variables to be roughly zero. In terms of magnitudes, the correlation coefficient has a maximum value of 1.0, or 100%, indicating perfect correlation (e.g., the temperature in Toronto as measured in Fahrenheit and Celsius), and a minimum value of 1.0, or 100%, indicating perfect negative correlation (e.g., the price of a zero-coupon bond and its yield). All values in between are valid, and the process lends itself to all the subjectivity that the human mind can muster. However, you may find the following (admittedly simplistic) rules of thumb to be useful: Value of Correlation Coefficient Less than 50% Between 50% and 10% Between 10% and 10% Between 10% and 50% Greater than 50% Interpretation High negative correlation—merits full investigation.

Correlations The final core element of our introductory statistical tool kit is correlation analysis. You ought to be at least nominally familiar with this concept, which involves identifying the extent to which two or more data series dynamically exhibit similar characteristics, most notably, for our purposes, across time. Correlation coefficients can range from 100% to 100% but (unless data series are simply disguised representations of a single concept, for example, the yield on a given bond and its price) typically fall somewhere in between. By performing correlation analysis on the time series of portfolio returns, traders stand to gain unique and specific insights into underlying portfolio economics.

Following is a summary of some of the standard categories of correlation analysis that you may find useful in identifying the drivers of relative performance for your portfolio. Correlation against Market Benchmarks. This is the general case associated with the example provided prior, under which you might calculate “correlation coefficients” between your returns and the performance 74 TRADING RISK of various market indexes. Here, I recommend that you begin the process by simply identifying, in an anecdotal sense, the market indexes that might best capture the essence of your trading and then running some introductory correlations there.

pages: 721 words: 197,134

Data Mining: Concepts, Models, Methods, and Algorithms by Mehmed Kantardzić

It is wrong to conclude that r = 0.6 indicates a linear relationship twice as strong as that indicated by the value r = 0.3. For our simple example of linear regression given at the beginning of this section, the model obtained was B = 0.8 + 0.92A. We may estimate the quality of the model using the correlation coefficient r as a measure. Based on the available data in Figure 4.3, we obtained intermediate results and the final correlation coefficient: A correlation coefficient r = 0.85 indicates a good linear relationship between two variables. Additional interpretation is possible. Because r2 = 0.72, we can say that approximately 72% of the variations in the values of B is accounted for by a linear relationship with A. 5.5 ANOVA Often the problem of analyzing the quality of the estimated regression line and the influence of the independent variables on the final regression is handled through an ANOVA approach.

Given a data set with two dimensions X and Y: X Y 1 5 4 2.75 3 3 5 2.5 (a) Use a linear regression method to calculate the parameters α and β where y = α + β x. (b) Estimate the quality of the model obtained in (a) using the correlation coefficient r. (c) Use an appropriate nonlinear transformation (one of those represented in Table 5.3) to improve regression results. What is the equation for a new, improved, and nonlinear model? Discuss a reduction of the correlation coefficient value. 5. A logit function, obtained through logistic regression, has the form: Find the probability of output values 0 and 1 for the following samples: (a) { 1, −1, −1 } (b) { −1, 1, 0 } (c) { 0, 0, 0 } 6.

On the other hand, the test for feature Y is significantly above the threshold value; this feature is not a candidate for reduction because it has the potential to be a distinguishing feature between two classes. A similar idea for feature ranking is shown in the algorithm that is based on correlation criteria. Let us consider first the prediction of a continuous outcome y. The Pearson correlation coefficient is defined as: where cov designates the covariance and var the variance. The estimate of R(i) for the given data set with samples’ inputs xk,,j and outputs yk is defined by: where the bar notation stands for an average over the index k (set of all samples). Using R(i)2 as a variable-ranking criterion enforces a ranking according to goodness of linear fit of individual variables.

pages: 130 words: 11,880

Optimization Methods in Finance by Gerard Cornuejols, Reha Tutuncu

Consider an investor who has a certain amount of money to be invested in a number of different securities (stocks, bonds, etc.) with random returns. For each security i, i = 1, . . . , n, estimates of its expected return, µi , and variance, σi2 , are given. Furthermore, for any two securities i and j, their correlation coefficient ρij is also assumed to be known. If we represent the proportion of the total funds invested in security i by xi , one can compute the expected return and the variance of the resulting portfolio x = (x1 , . . . , xn ) as follows: E[x] = x1 µ1 + . . . + xn µn = µT x, and V ar[x] = X ρij σi σj xi xj = xT Qx i,j where ρii ≡ 1, Qij = ρij σi σj for i 6= j, Qii = σi2 , and µ = (µ1 , . . . , µn ).

Chapter 5 QP Models and Tools in Finance 5.1 Mean-Variance Optimization In the introductory chapter, we have discussed Markowitz’ theory of mean-variance optimization (MVO) for the selection of portfolios of securities (or asset classes) in a manner that trades off the expected returns and the perceived risk of potential portfolios. Consider assets S1 , S2 , . . . , Sn (n ≥ 2) with random returns. Let µi and σi denote the expected return and the standard deviation of the return of asset Si . For i 6= j, ρij denotes the correlation coefficient of the returns of assets Si and Sj . Let µ = [µ1 , . . . , µn ]T , and Q be the n × n symmetric covariance matrix with Qii = σi2 and Qij = ρij σi σj for i 6= j. Denoting the proportion of the total funds invested in security i by xi , one can represent the expected return and the variance of the resulting portfolio x = (x1 , . . . , xn ) as follows: E[x] = x1 µ1 + . . . + xn µn = µT x, and V ar[x] = X ρij σi σj xi xj = xT Qx, i,j where ρii ≡ 1.

The standard deviation of a random variable is the square-root of its variance. 108 APPENDIX C. A PROBABILITY PRIMER For two jointly distributed random variables X1 and X2 , their covariance is defined to be Cov(X1 , X2 ) = E [(X1 − E[X1 ])(X2 − E[X2 ])] = E[X1 X2 ] − E[X1 ]E[X2 ] The correlation coefficient of two random variables is the ratio of their covariance to the product of their standard deviations. For a collection of random variables X1 , . . . , Xn , the expected value of the sum of these random variables is equal to the sum of their expected values: " E n X i=1 # Xi = n X E[Xi ].

The Art of Computer Programming by Donald Ervin Knuth

We may also compute the following statistic: This is the "serial correlation coefficient," a measure of the extent to which Uj+i depends on Uj. Correlation coefficients appear frequently in statistical work. If we have n quantities Uo, Ui, ..., t/n-i and n others Vo, Vi, ..., Vn_i, the correlation coefficient between them is defined to be c = All summations in this formula are to be taken over the range 0 < j < n; Eq. B3) is the special case Vj = C/(j+i) mod n- The denominator of B4) is zero when JJo = U\ = ¦ ¦ ¦ = Un-\ or V\$ = V\ = ¦ ¦ ¦ = Vn-\\ we exclude that case from discussion. 3.3.2 EMPIRICAL TESTS 73 A correlation coefficient always lies between —1 and +1.

Show that the correlation coefficient C given in Eq. B4) is equal to ? u'kvL 0<k<n b) Let C = N/D, where N and D denote the numerator and denominator of the expression in part (a). Show that N2 < D2, hence — 1 < C < 1; and obtain a formula for the difference D2 - N2. [Hint: See exercise 1.2.3-30.] c) If C = ±1, show that aUk + CVk = t, 0 < k < n, for some constants a, C, and r, not all zero. 18. [M20] (a) Show that if n = 2, the serial correlation coefficient B3) is always equal to —1 (unless the denominator is zero), (b) Similarly, show that when n = 3, the serial correlation coefficient always equals — \.

In actual fact, since U0U1 is not completely independent of U1U2, the serial correlation coefficient is not expected to be exactly zero. (See exercise 18.) A "good" value of C will be between \xn — 2an and \xn + 2an, where l" ">2' B5) "-=;rrr °l=(n-1) (n - 2)' ">2' We expect C to be between these limits about 95 percent of the time. The formula for a\ in B5) is an upper bound, valid for serial correlations between independent random variables from an arbitrary distribution. When the C/'s are uniformly distributed, the true variance is obtained by subtracting %r-n~2 + O(n~7/3 logn). (See exercise 20.) Instead of simply computing the correlation coefficient between the obser- observations (Uo, U\, ..., Un-\) and their immediate successors (U\,..., Un-i,Uo), we can also compute it between (Uo, U\,..., Un-\) and any cyclically shifted sequence (Uq,...

pages: 353 words: 148,895

Triumph of the Optimists: 101 Years of Global Investment Returns by Elroy Dimson, Paul Marsh, Mike Staunton

The inter-war period, with its post-war boom, hyperinflation in Germany, the Wall Street Crash, and the Great Depression, was unique. Correlations were quite high due to common factors such as the crash and Depression, but the correlation structure differed from all other periods. Figure 8-6: Correlation coefficients between four core countries over seven successive sub-periods 0.7 0.6 0.5 0.4 Correlation coefficients US:UK US:Fra UK:Fra Average US:Ger UK:Ger Ger:Fra .40 .26 0.3 0.2 0.1 .09 .14 .15 .01 0.0 -0.1 -.07 -0.2 -0.3 -0.4 -0.5 1872–1889 1889–1914 Source: Goetzmann, Li, and Rouwenhorst, 2001 1915–1918 1919–1939 1940–1945 1946–1971 1972–2000 Chapter 8: International investment 117 Longin and Solnik (1995) provide further evidence of high correlations during periods of poor performance.

France’s highest correlations were with Belgium, The Netherlands, Italy, Ireland, Spain, and Switzerland; Italy’s were with France and Switzerland; The Netherlands was most highly correlated with Belgium, followed by France, Denmark, and Switzerland; and Sweden was highly correlated with Denmark, Canada (natural resources), and Switzerland (neutral countries). Australia’s highest correlations were with the United Kingdom and Ireland (historical and trade links), and Canada and South Africa (gold, mining, and the British Empire). 115 Chapter 8: International investment Table 8-3: Correlation coefficients between world equity markets* Wld Wld US UK .93 Swi Swe Spa SAf Neth Jap Ita Ire Ger Fra Den Can Bel Aus .77 .59 .62 .67 .54 .73 .68 .52 .69 .69 .73 .57 .82 .54 .69 .67 .44 .46 .53 .46 .57 .49 .40 .66 .56 .56 .46 .78 .45 .57 US .85 UK .70 .55 Swi .68 .50 .62 Swe .62 .44 .42 .54 Spa .41 .25 .25 .36 .37 SAf .55 .43 .49 .39 .34 .26 Neth .57 .39 .42 .51 .43 .28 .58 .44 .63 .31 .71 .42 .39 .73 .58 .59 .57 .57 .59 .56 .39 .60 .19 .72 .36 .45 .57 .53 .64 .58 .35 .63 .37 .63 .38 .63 .34 .49 .27 .76 .76 .44 .61 .29 .44 .35 .63 .32 .64 .50 .64 .75 .56 .51 .55 .54 .30 .29 .44 .24 .31 .42 .37 .25 .62 .10 .66 .39 .59 .63 .74 .77 .64 .55 .70 .46 Jap .45 .21 .33 .29 .39 .40 .31 .25 Ita .54 .37 .43 .52 .39 .41 .41 .32 Ire .58 .38 .73 .70 .42 .35 .42 .46 .29 .43 Ger .30 .12 -.01 .22 .09 -.03 .05 .27 .06 .16 .18 .34 .33 .25 .36 .24 .50 .17 .59 .33 .55 .71 .50 .40 .51 .38 .42 .03 .45 .49 .54 .57 .50 .83 .61 .57 .59 .46 Fra .62 .36 .45 .54 .44 .47 .38 .48 .25 .52 .53 .19 Den .57 .38 .40 .51 .56 .34 .31 .50 .46 .38 .55 .22 .45 Can .80 .80 .55 .48 .53 .27 .54 .34 .30 .37 .41 .13 .35 .46 Bel .58 .38 .40 .57 .43 .40 .29 .60 .25 .47 .49 .26 .68 .42 .35 Aus .66 .47 .66 .51 .50 .28 .56 .41 .28 .43 .62 .04 .47 .42 .62 .63 .60 .66 .48 .55 .54 .30 .30 .65 .30 .35 * Correlations in bold (lower left-hand triangle) are based on 101 years of real dollar returns, 1900–2000.

Correlations have nevertheless shifted significantly over time, and Table 8-4 provides evidence of this. The table presents the results of estimating correlations over a prediction period from information over an earlier historical period. The top panel shows that when the full set of pairwise correlation coefficients between equity markets are estimated separately for the first and second halves of the twentieth century, there was no discernable relationship between the two. It would not have been possible to predict correlations for 1950–2000 from those estimated from annual data over the first half-century.

Principles of Corporate Finance by Richard A. Brealey, Stewart C. Myers, Franklin Allen

As you might guess, the covariance is a measure of the degree to which the two stocks “covary.” The covariance can be expressed as the product of the correlation coefficient ρ12 and the two standard deviations:28 For the most part stocks tend to move together. In this case the correlation coefficient ρ12 is positive, and therefore the covariance σ12 is also positive. If the prospects of the stocks were wholly unrelated, both the correlation coefficient and the covariance would be zero; and if the stocks tended to move in opposite directions, the correlation coefficient and the covariance would be negative. Just as you weighted the variances by the square of the proportion invested, so you must weight the covariance by the product of the two proportionate holdings x1 and x2.

Recalculate the expected portfolio return and standard deviation for different values of x1 and x2, assuming the correlation coefficient ρ12 = 0. Plot the range of possible combinations of expected return and standard deviation as in Figure 8.3. Repeat the problem for ρ12 = +.25. 11. Portfolio risk and return Mark Harrywitz proposes to invest in two shares, X and Y. He expects a return of 12% from X and 8% from Y. The standard deviation of returns is 8% for X and 5% for Y. The correlation coefficient between the returns is .2. a. Compute the expected return and standard deviation of the following portfolios: b.

Calculate the variance and standard deviation of the returns on a portfolio that has equal investments in 2 shares, 3 shares, and so on, up to 10 shares. b. Use your estimates to draw a graph like Figure 7.11. How large is the underlying market risk that cannot be diversified away? c. Now repeat the problem, assuming that the correlation between each pair of stocks is zero. 17. Portfolio risk Table 7.9 shows standard deviations and correlation coefficients for eight stocks from different countries. Calculate the variance of a portfolio with equal investments in each stock. 18. Portfolio risk Your eccentric Aunt Claudia has left you \$50,000 in BP shares plus \$50,000 cash. Unfortunately her will requires that the BP stock not be sold for one year and the \$50,000 cash must be entirely invested in one of the stocks shown in Table 7.9.

pages: 315 words: 93,628

Is God a Mathematician? by Mario Livio

For a given value of the temperature, one cannot predict precisely the number of forest fires that will break out, since the latter depends on other variables such as the humidity and the number of fires started by people. In other words, for any value of the temperature, there could be many corresponding numbers of forest fires and vice versa. Still, the mathematical concept known as the correlation coefficient allows us to measure quantitatively the strength of the relationship between two such variables. The person who first introduced the tool of the correlation coefficient was the Victorian geographer, meteorologist, anthropologist, and statistician Sir Francis Galton (1822–1911). Galton—who was, by the way, the half-cousin of Charles Darwin—was not a professional mathematician.

If the correlation between them is very close, a very long cubit would usually imply a very tall stature, but if it were not very close, a very long cubit would be on the average associated with only a tall stature, and not a very tall one; while, if it were nil, a very long cubit would be associated with no especial stature, and therefore, on the average, with mediocrity. Pearson eventually gave a precise mathematical definition of the correlation coefficient. The coefficient is defined in such a way that when the correlation is very high—that is, when one variable closely follows the up-and-down trends of the other—the coefficient takes the value of 1. When two quantities are anticorrelated, meaning that when one increases the other decreases and vice versa, the coefficient is equal to–1. Two variables that each behave as if the other didn’t even exist have a correlation coefficient of 0. (For instance, the behavior of some governments unfortunately shows almost zero correlation with the wishes of the people whom they supposedly represent.)

Data Mining: Concepts and Techniques: Concepts and Techniques by Jiawei Han, Micheline Kamber, Jian Pei

Since our computed value is above this, we can reject the hypothesis that gender and preferred_reading are independent and conclude that the two attributes are (strongly) correlated for the given group of people. Correlation Coefficient for Numeric Data For numeric attributes, we can evaluate the correlation between two attributes, A and B, by computing the correlation coefficient (also known as Pearson's product moment coefficient, named after its inventer, Karl Pearson). This is(3.3) where n is the number of tuples, ai and bi are the respective values of A and B in tuple i, Ā and are the respective mean values of A and B, σA and σB are the respective standard deviations of A and B (as defined in Section 2.2.2), and Σ(aibi) is the sum of the AB cross-product (i.e., for each tuple, the value for A is multiplied by the value for B in that tuple).

Inconsistencies in attribute or dimension naming can also cause redundancies in the resulting data set. Some redundancies can be detected by correlation analysis. Given two attributes, such analysis can measure how strongly one attribute implies the other, based on the available data. For nominal data, we use the χ2 (chi-square) test. For numeric attributes, we can use the correlation coefficient and covariance, both of which access how one attribute's values vary from those of another. χ2 Correlation Test for Nominal Data For nominal data, a correlation relationship between two attributes, A and B, can be discovered by a χ2 (chi-square) test. Suppose A has c distinct values, namely a1, a2, … ac.

Consider two numeric attributes A and B, and a set of n observations {(a1, b1), …, (an, bn)}. The mean values of A and B, respectively, are also known as the expected values on A and B, that is, and The covariance between A and B is defined as(3.4) If we compare Eq. (3.3) for rA, B (correlation coefficient) with Eq. (3.4) for covariance, we see that(3.5) where σA and σB are the standard deviations of A and B, respectively. It can also be shown that(3.6) This equation may simplify calculations. For two attributes A and B that tend to change together, if A is larger than Ā (the expected value of A), then B is likely to be larger than (the expected value of B).

The Concepts and Practice of Mathematical Finance by Mark S. Joshi

In particular, we have that E((Yt - YS)(Xil) - X(1))) = PE((Xtl) - Xs 1))2) + 1 - p2]E((Xi2)- X(Z))(Xt`1 Xs1))). (11.4) Since V) and X(2) are independent, the second expectation is zero and so 1E((Yt - (Xrl) YS) - Xs 1))) = p(t - s). (11.5) As Yt - YS and Xtl) - XSl) both have variance t - s, this means that the correlation coefficient is p. Thus we have constructed a Brownian motion whose increments are correlated to those of X(1) with correlation p. More generally, we could construct a Brownian motion from any vector a = (al,ak) with a? = 1, by taking Ek=l ajX(J). 11.3 The higher-dimensional Ito calculus 263 The existence of such correlated Brownian motions will be crucial in pricing multi-asset options.

(11.12) The second term has mean zero and variance of order At2 so we can discard it as small, whereas the first term has mean pjk At and variance of order At2 and therefore contributes. This gives us a new rule for the multi-dimensional Ito calculus: dWrj)dW(k) = pjkdt. To summarize, we have Theorem 11.1 (Multi-dimensional Ito lemma) Let Wtj) be correlated Brownian motions with correlation coefficient pjk between the Brownian motions WU) and Wtk). Let Xj be an Ito process with respect to Wt W. Let f be a smooth function; we then have that af of 11 at j=1 11 1 T a2f + 2 j,k=1 ax axi` (t, X1, ax j (t,X1,...,X,1)dXj ..., X7,)dX jdXk, (11.14) with dWtj)dWtk) = pjkdt. When collecting terms, the final double sum will be absorbed into the dt term.

Perfect correlation means the vectors point the same way, perfect negative correlation means they point the opposite way, and zero correlation means they are orthogonal. In (11.20), the first vector has length O'1, and the second length oa2. When we add two vectors, v1, V2, the square of the length of the resultant vector is IIv1112 + 2 cos(9)IIv111. i1v211 + IIv2112, where 9 is the angle between the vectors. If we interpret the correlation coefficient as being the cosine of the angle between the two Brownian motions, then this means that the new volatility is just the length of the vector obtained by summing the vectors for each Brownian motion. 266 Multiple sources of risk More generally, we could construct a Brownian motion from any vector a=(a1,...

The Book of Why: The New Science of Cause and Effect by Judea Pearl, Dana Mackenzie

Galton’s disciple Karl Pearson later derived a formula for the slope of the (properly rescaled) regression line and called it the correlation coefficient. This is still the first number that statisticians all over the world compute when they want to know how strongly two different variables in a data set are related. Galton and Pearson must have been thrilled to find such a universal way of describing the relationships between random variables. For Pearson, especially, the slippery old concepts of cause and effect seemed outdated and unscientific, compared to the mathematically clear and precise concept of a correlation coefficient. GALTON AND THE ABANDONED QUEST It is an irony of history that Galton started out in search of causation and ended up discovering correlation, a relationship that is oblivious of causation.

It was Galton who first freed me from the prejudice that sound mathematics could only be applied to natural phenomena under the category of causation.” In Pearson’s eyes, Galton had enlarged the vocabulary of science. Causation was reduced to nothing more than a special case of correlation (namely, the case where the correlation coefficient is 1 or –1 and the relationship between x and y is deterministic). He expresses his view of causation with great clarity in The Grammar of Science (1892): “That a certain sequence has occurred and reoccurred in the past is a matter of experience to which we give expression in the concept causation.… Science in no case can demonstrate any inherent necessity in a sequence, nor prove with absolute certainty that it must be repeated.”

The parameter a (often denoted by rYX, the regression coefficient of Y on X) tells us the average observed trend: a one-unit increase of X will, on average, produce an a-unit increase in Y. If there are no confounders of Y and X, then we can use this as our estimate of an intervention to increase X by one unit. But what if there is a confounder, Z? In this case, the correlation coefficient rYX will not give us the average causal effect; it only gives us the average observed trend. That was the case in Wright’s problem of the guinea pig birth weights, discussed in Chapter 2, where the apparent benefit (5.66 grams) of an extra day’s gestation was biased because it was confounded with the effect of a smaller litter size.

pages: 206 words: 70,924

The Rise of the Quants: Marschak, Sharpe, Black, Scholes and Merton by Colin Read

These assumptions allow us to calculate the expected return of the portfolio Rp by summing across all securities: E ( Rp ) = ∑ wi E (Ri ) i and the portfolio variance: s 2p = ∑ wi2 s i2 + ∑ ∑ wi w j si sj r ij i i j ≠i Notice that all of the coefficients on the right-hand side of the portfolio variance expression are necessarily positive, except for the correlation coefficient. We can readily see that portfolio variance is minimized if the correlation coefficient ij 1. We can generalize this risk minimization procedure through the matrix algebra for which Markowitz developed efficient solution algorithms that were more easily computable. This matrix algebra approach that minimizes variance for a given return R and wealth w becomes: min s 2 = min XVX T ∋ r = (W − X.1)rf + XR X X where the wealth constraint r = ( W − X.1) rf + XR affirms that wealth is invested in a risky portfolio R that returns R and a risk-free asset that returns rf.

Most significantly for financial pricing theory, he went on: [W]e reinterpret [the decision variables] to mean not future yields but parameters [e.g., moments and joint moments] of the jointfrequency distribution of future yields. Thus, x may be interpreted as the mathematical expectation of first year’s meat consumption, y may be its standard deviation, z may be the correlation coefficient between meat and salt consumption … etc. … It is sufficiently realistic, however, to confine ourselves, for each [return] to two parameters only: the mathematical expectation … and the coefficient of variation [“risk”].8 Marschak proposed a simple approach to the consideration of the interplay between return and risk by confining its description to first moments, known as means, and second moments of returns, labeled variances and covariances.

pages: 757 words: 193,541

The Practice of Cloud System Administration: DevOps and SRE Practices for Web Services, Volume 2 by Thomas A. Limoncelli, Strata R. Chalup, Christina J. Hogan

Headroom is usually specified as a percentage of current capacity. • Timetable: For each component, what is the lead time from ordering to delivery, and from delivery until it is in service? Are there specific constraints for bringing new capacity into service, such as change windows? * * * Math Terms Correlation Coefficient: Describes how strongly measurements for different data sources resemble each other. Moving Average: A series of averages, each of which is taken across a short time interval (window), rather than across the whole data set. Regression Analysis: A statistical method for analyzing relationships between different data sources to determine how well they correlate, and to predict changes in one based on changes in another.

The number of data samples in that time period is n. If your core driver metric is x and your primary resource metric is y, you first calculate the sum of the last n values for x, x2, y, y2, and x times y, giving Σx, Σx2, Σy, Σy2, and Σxy. Then calculate SSxy, SSxx, SSyy, and R as follows: Regression analysis results in a correlation coefficient R, which is a number between –1 and 1. Squaring this number and then multiplying by 100 gives the percentage match between the two data sources. For example, for the MAU and network utilization figures shown in Figure 18.2, this calculation gives a very high correlation, between 96 percent and 100 percent, as shown in Figure 18.3, where R2 is graphed.

The large fluctuations in b for the length of the correlation window are due to significant changes in the moving averages from day to day, as the moving average has both pre- and post-upgrade data. When sufficient time has passed so that only post-upgrade data is used in the moving average, b becomes stable and the correlation coefficient returns to its previous high levels. The value of b corresponds to the slope of the line, or the multiplier in the equation linking the core driver and the usage of the primary resource. When correlation returns to normal, b is at a higher level. This result indicates that the primary resource will be consumed more rapidly with this software release than with the previous one.

pages: 586 words: 159,901

Wall Street: How It Works And for Whom by Doug Henwood

Sometimes it doesn't matter whether the bad news is true; if the short can take a position and undertake a successful disinformation campaign, he or she can profitably cover the short. 35. For the "real" sector, however, borders still matter, and the "global assembly line" is a bit of an exaggeration. 36. The correlation coefficient is a measure of how tightly two sets of numbers are related to each other, ranging from -1 (a perfect mirror image) through 0 (no relation at all) to PLAYERS +1 (perfect lockstep). A correlation coefficient under 0.2 marks a fairly cacophanous relation, but figures over 0.9 signify great intimacy. 37. In fact, many foreign investments made in the U.S. during the 1980s have had apparently dismal rates of return.

Most of the sovereign defaulters, by the way, had good ratings from Moody's (Cantor and Packer 1995). But now those defaults are a distant memory, and today's capital markets look seamless. Statistics confirm the decreasing importance of borders for the financial markets."*^ In the 1970s, the correlation coefficient between interest rates on 10-year U.S. government bonds and German bonds of similar maturity was 0.191, but from 1990 to 1994, it was 0.934; Japan and the U.S., 0.182 and 0.965, respectively; and the U.S. and the U.K., 0.590 and 0.949 (Bank of England data, reported in Goldstein et al. 1994, p. 5).-^'' While it would be an exaggeration to say that there's now a single global credit market, we're definitely moving in that direction.

One doesn't want to get too carried away naturalizing temperament and values, but the model seems particularly to drive away women and nonwhites, at least in America, because of its chilly irreality. It may just be that sex and race are simply convenient markers for hierarchy —that economics is an ideology of privilege, and the already privileged, or those who wish to become apologists for the privileged, are drawn to its study. WALL STREET 8. Correlation coefficients for the various versions of q suggested in the text are all well over .92. The correlation for the simple equity q (market value of stock divided by tangible assets, as shown in the charts and used in the text) and the values for 1960-74 reported in Tobin and Brainard C1977) is .97. 9- It's interesting that investment rose during what are usually considered the bad years of the 1970s.

pages: 467 words: 154,960

Trend Following: How Great Traders Make Millions in Up or Down Markets by Michael W. Covel

They traded with their gut and treated drawdowns as a cancer, rather than the natural ebb and flow of trading.” Interestingly, there is another perspective on drawdowns that few people consider. When you look at trend following performance data—for example, Dunn’s track record—you can’t help but notice that certain times are better than others to invest with Dunn. Correlation coefficient: A statistical measure of the interdependence of two or more random variables. Fundamentally, the value indicates how much of a change in one variable is explained by a change in another.25 Smart clients of Dunn look at his performance chart and buy in when his fund is experiencing a drawdown.

Unlike misguided comparisons, such as using standard deviation, I find correlation comparisons of performance data useful. In a research paper titled “Learning to Love Non-Correlation,” correlation is defined as “a statistical term giving the strength of linear relationship between two random variables. It is the historical tendency of one thing to move in tandem with another.” The correlation coefficient is a number from –1 to +1, with –1 being the perfectly opposite behavior of two investments (for example, up 5 percent every time the other is down 5 percent). The +1 reflects identical investment results (up or down the same amount each period). The further away from +1 one gets (and thus closer to –1), the better a diversifier one investment is for the other.

The further away from +1 one gets (and thus closer to –1), the better a diversifier one investment is for the other. But because his firm is keenly aware of keeping things simple, it also provides another description of correlation: the tendency for one investment to “zig” while another “zags.”27 I took the monthly performance numbers of trend followers and computed their correlation coefficients. Comparing correlations provided evidence that trend followers trade typically the same markets in the same way at the same time. Look at the correlation chart (see Chart 3.5) and ask yourself: “Why do two trend followers who don’t work in the same office, who are on opposite sides of the continent, have the same three losing months in a row with similar percentage losses?”

The Handbook of Personal Wealth Management by Reuvid, Jonathan.

Forestry in the United States has been repeatedly shown to have a negative correlation coefficient with, among other financial assets, common stocks, corporate and government bonds, and the S&P 500 (see Table 2.3.1), and in certain studies reduced real portfolio risk by an average of 5 per cent.1 We observe that forestry generally forms a minor element of an overall investment portfolio, perhaps no more than 5–10 per cent as a maximum. Table 2.3.1 Timberland correlation coefficients, 1959–78 Investment correlation coefficient Timberland Residential housing Farm real estate S&P 500 index OTC stocks Preferred stock average No-load mutual fund average Municipal bonds Treasury Bills Long-term corporate bonds Commodity futures average 1.0000 –0.0905 0.5612 –0.4889 –0.4917 –0.3533 –0.6351 –0.0900 0.3118 –0.2704 0.8988 Source: Zinkhan, FC, Sizemore, WR, Mason, GH and Ebner, TJ (1992) Timberland Investments, Portland, Oregon, Timber Press. _______________________________________ CURRENT OPPORTUNITIES IN FORESTRY 81 ឣ Liquidity Certain forestry investments such as COEIC funds and exchange-traded funds (see below) are traded daily on markets such as the London Stock Exchange and Alternative Investment Market (AIM).

pages: 147 words: 42,682

Facing Reality: Two Truths About Race in America by Charles Murray

You will find brilliant performers of every race in any occupation. That doesn’t negate the relevance of these considerations to group means. The magnitude of the relationship of cognitive ability to job performance varies. Magnitude in this case is usually expressed as the correlation coefficient between IQ and a measure of job performance. A correlation coefficient goes from –1 (a perfectly inverse relationship) to +1 (a perfectly direct relationship). The square of a correlation represents the percentage of the variance it “explains.” Rules of thumb are that the correlations between IQ scores and job productivity for low-complexity jobs are seldom below .2; for medium-complexity jobs, they are seldom below .4; for high-complexity jobs, they are seldom below .5.

pages: 266 words: 86,324

The Drunkard's Walk: How Randomness Rules Our Lives by Leonard Mlodinow

For example, if data revealed that by eating the latest McDonald’s 1,000-calorie meal once a week, people gained 10 pounds a year and by eating it twice a week they gained 20 pounds, and so on, the correlation coefficient would be 1. If for some reason everyone were to instead lose those amounts of weight, the correlation coefficient would be -1. And if the weight gain and loss were all over the map and didn’t depend on meal consumption, the coefficient would be 0. Today correlation coefficients are among the most widely employed concepts in statistics. They are used to assess such relationships as those between the number of cigarettes smoked and the incidence of cancer, the distance of stars from Earth and the speed with which they are moving away from our planet, and the scores students achieve on standardized tests and the income of the students’ families.

pages: 363 words: 28,546

Portfolio Design: A Modern Approach to Asset Allocation by R. Marston

But for the last 10 years alone ending in 2009, the correlation between EAFE and the S&P rises to 0.87. There are correspondingly large increases in correlations between the S&P and the regional MSCI indexes. When did this increase in correlations occur? Consider Figure 5.9 which shows five- and 10-year correlation coefficients between the EAFE and S&P 500 indexes. Since the EAFE index starts only in 1970, the graph begins in 1975 for the five-year correlation and in 1980 for the 10-year correlation. The figure is noteworthy in several respects. First, the correlations vary widely over time whether they are measured over five- or ten-year intervals.

The Russell 1000 Growth Index is dominated by the Russell 1000 Value Index. This result should not be surprising given the analysis in Chapter 4. Russell 1000 Value has a higher return and a lower standard deviation than Russell 1000 Growth. What’s more, the two indexes are highly correlated with a correlation coefficient of 0.82. The optimizer finds that one series is totally dominated by the other. So the optimizer rejects one whole asset class. The optimizer is also not fond of small-cap stocks. The Russell 2000 Index has a small weighting in the lowest risk portfolios, and its role disappears in portfolios with larger allocations to stocks.

The overlap among the three largest databases was analyzed after eliminating the funds that only appeared in MSCI. The percentages were rounded to the nearest decimal. 12. As explained in Chapter 5, the underperformance of EAFE relative to U.S. stocks is almost entirely due to Japan. 13. Recall that beta is equal to the correlation coefficient times the ratio of the standard deviation of the asset relative to the standard deviation of the benchmark. The beta is 0.36 = 0.77 ∗ (0.071/0.152). 14. Since the average return on the risk-free Treasury bill is 3.8 percent and the average return on the Russell 3000 is 9.2 percent, the alpha = 11.8 percent – [3.8% + 0.36∗(9.2% – 3.8%)] = 6.1%. 15.

pages: 545 words: 137,789

How Markets Fail: The Logic of Economic Calamities by John Cassidy

As the big hedge fund Long-Term Capital Management discovered to its cost during the international financial crisis of 1998, many assets that seem to have little or nothing in common suddenly move in the same direction. Prior to the blowup, for example, the correlation coefficient between certain bonds issued by the governments of the Philippines and Bulgaria was just 0.04: as the crisis unfolded, their correlation coefficient rose to 0.84. (A correlation coefficient of zero means two assets have no relationship; a coefficient of one means they move in perfect unison.) During a period of market upheaval, as a Wall Street saying has it, “all correlations go to one.”

. : Philippe Jorion, Value at Risk: The New Benchmark for Managing Financial Risk, 2nd ed. (New York: McGraw-Hill, 2000), 107. 274 “In contrast with traditional . . .”: Ibid., xxii. 275 “It helps you understand . . .”: Quoted in Joe Nocera, “Risk Mismanagement,” New York Times Magazine, January 2, 2009. 277 the correlation coefficient . . . : Linda Allen, Jacob Boudoukh, and Anthony Saunders, Understanding Market, Credit, and Operational Risk: The Value at Risk Approach (Hoboken, N.J.: Wiley-Blackwell, 2004), 103. 278 “We remind our readers . . .”: “CreditMetrics Technical Document,” RiskMetrics, April 1997, available at www.riskmetrics.com/publications/techdocs/cmtdovv.html. 278 “The relative prevalence of . . .”: Allen et al., Understanding Market, 35. 278 “I believe that . . .”: “Against Value at Risk: Nassim Taleb Replies to Philippe Jorion,” 1997, available at www.fooledbyrandomness.com/jorion.html. 279 “business planning relied on . . .”: UBS, “Shareholder Report on UBS’s Write-Downs,” 34. 279 “even though delinquency . . .”: Ibid., 38–39. 280 “would overturn . . .”: Gillian Tett, Fool’s Gold: How the Bold Dream of a Small Tribe at J.P.

pages: 335 words: 94,657

The Bogleheads' Guide to Investing by Taylor Larimore, Michael Leboeuf, Mel Lindauer

While some bond funds invest in highly rated investment-grade bonds, still others invest in lower-rated junk bonds. For more information on the various types of bonds, see Chapter 3. When investments (like stocks and bonds) don't always move together, they're said to have a low correlation coefficient. Understanding the correlation coefficient principal isn't really that difficult. The correlation numbers for any two investments can range from +1.0 (perfect correlation) to -1.0 (negative correlation) Basically, if two stocks (or funds) normally move together at the same rate, they're said to be highly correlated, and when two investments move in the opposite directions, they're said to be negatively correlated.

When two investments each randomly go their separate ways, independent of the movement of the other one, there is said to be no correlation between them, and their correlation figure would be shown as 0. Finally, when two investments always move in the opposite direction, they would have a negative correlation, which would be represented by a rating of -1.0. In actual practice, you'll find that most investment choices available to you will have a correlation coefficient somewhere between 1.0 (perfect correlation) and 0 (noncorrelated). It's very difficult to find negatively correlated asset classes that have similar expected returns. The closer the number is to 1.0, the higher the correlation between the two assets, and the lower the number, the less correlation there is between the two investments.

pages: 654 words: 191,864

Thinking, Fast and Slow by Daniel Kahneman

It was a simple matter to rank the advisers by their performance in each year and to determine whether there were persistent differences in skill among them and whether the same advisers consistently achieved better returns for their clients year after year. To answer the question, I computed correlation coefficients between the rankings in each pair of years: year 1 with year 2, year 1 with year 3, and so on up through year 7 with year 8. That yielded 28 correlation coefficients, one for each pair of years. I knew the theory and was prepared to find weak evidence of persistence of skill. Still, I was surprised to find that the average of the 28 correlations was .01. In other words, zero.

If all you know about Tom is that he ranks twelfth in weight (well above average), you can infer (statistically) that he is probably older than average and also that he probably consumes more ice cream than other children. If all you know about Barbara is that she is eighty-fifth in piano (far below the average of the group), you can infer that she is likely to be young and that she is likely to practice less than most other children. The correlation coefficient between two measures, which varies between 0 and 1, is a measure of the relative weight of the factors they share. For example, we all share half our genes with each of our parents, and for traits in which environmental factors have relatively little influence, such as height, the correlation between parent and child is not far from .50.

In one study, the CEOs were characterized by the strategy of the companies they had led before their current appointment, as well as by management rules and procedures adopted after their appointment. CEOs do influence performance, but the effects are much smaller than a reading of the business press suggests. Researchers measure the strength of relationships by a correlation coefficient, which varies between 0 and 1. The coefficient was defined earlier (in relation to regression to the mean) by the extent to which two measures are determined by shared factors. A very generous estimate of the correlation between the success of the firm and the quality of its CEO might be as high as .30, indicating 30% overlap.

pages: 193 words: 47,808

The Flat White Economy by Douglas McWilliams

b) Growing Together: London and the UK Economy 2005 This report looks at a range of links between the London economy and those of the rest of the UK.18 It concludes that London’s growth is not at the expense of the rest of the UK, but that London and other UK regions and countries are interdependent. Table 6.1: Correlation between economic growth in London and the rest of the UK, 1983–2004 Regions and countries of Great Britain Correlation coefficient South East 0.80 East England 0.81 South West 0.64 East Midlands 0.45 West Midlands 0.73 North West 0.73 Yorkshire and Humberside 0.56 North East 0.22 Wales 0.55 Scotland 0.27 Northern Ireland 0.36 Table 6.2: Percentage change in employment, 1989–2001 Regions and countries of Great Britain % change South East 23.7 South West 21.2 East of England 18.8 Scotland 17.4 London 15.3 East Midlands 12.5 Wales 11.7 West Midlands 10.8 Yorkshire and Humberside 10.2 North West 9.9 North East 7.2 A unique feature of this analysis is research into the correlations between GVA growth in London and other regions and countries.

It is also interesting that the three emerging economies for which we have data in this sample – China, Mexico and South Korea – have much lower labour shares of income than in the advanced economies (which might be a bit of evidence to support Marx’s contention that capitalism in the long term would ultimately bid profits down to a level that is too low to permit economic growth). However, these emerging economies currently have much faster rates of economic growth. What is the evidence that higher profits boost economic growth? A crude statistical analysis for the OECD economies for which data is available shows a negative correlation coefficient of -0.31 between the labour income share average from 2000 to 2006 and the rate of economic growth from 2001 to 2008. What this says is that there is a statistically significant negative correlation between the labour income share and GDP growth. In other words, the higher the share of profits, the faster the rate of growth.

All About Asset Allocation, Second Edition by Richard Ferri

These relationships are dynamic, and they can and do change without warning. Selecting investments that do not go up and down at the same time (or most of the time) can be made easier with correlation analysis. This is a mathematical measure of the tendency of one investment to move in relation to another. The correlation coefficient is a mathematically derived number that measures this tendency toward comovement relative to the investments’ average return. If two investments each move in the same direction at the same time above their average returns, they have a positive correlation. If they each move in opposite directions below their average returns, they have a negative correlation.

These pairs of investments just do not exist. Correlation is measured using a range between ⫹1 and ⫺1. Two investments that have a correlation of ⫹0.3 or greater are considered positively correlated. When two investments have a correlation of ⫺0.3 or less, this is considered negative correlation. A correlation coefficient between ⫺0.3 and ⫹0.3 is considered noncorrelated. When two investments are noncorrelated, either the movement of one does not track the movement of the other or the tracking is inconsistent and shifts between positive and negative. Figure 3-4 represents two investments that are noncorrelated; sometimes they move together, and sometimes they do not.

Closed-End Fund A mutual fund that has a fixed number of shares, usually listed on a major stock exchange. Commodities Unprocessed goods, such as grains, metals, and minerals, traded in large amounts on a commodities exchange. Consumer Price Index (CPI) A measure of the price change in consumer goods and services. The CPI is used to track the pace of inflation. Correlation Coefficient A number between ⫺1 and 1 that measures the degree to which two variables are linearly related. Cost Basis The original cost of an investment. For tax purposes, the cost basis is subtracted from the sale price to determine any capital gain or loss. Glossary 323 Country Risk The possibility that political events (e.g., a war, national elections), financial problems (e.g., rising inflation, government default), or natural disasters (e.g., an earthquake, a poor harvest) will weaken a country’s economy and cause investments in that country to decline.

pages: 764 words: 261,694

The Elements of Statistical Learning (Springer Series in Statistics) by Trevor Hastie, Robert Tibshirani, Jerome Friedman

This is the covariance matrix between these two variables, after linear adjustment for all the rest. In the Gaussian distribution, this is the covariance matrix of the conditional distribution of Xa |Xb . The partial correlation coefficient ρjk|rest between the pair Xa conditional on the rest Xb , is simply computed from this partial covariance. Define Θ = Σ−1 . 1. Show that Σa.b = Θ−1 aa . 2. Show that if any off-diagonal element of Θ is zero, then the partial correlation coefficient between the corresponding variables is zero. 3. Show that if we treat Θ as if it were a covariance matrix, and compute the corresponding “correlation” matrix R = diag(Θ)−1/2 · Θ · diag(Θ)−1/2 , then rjk = −ρjk|rest Ex. 17.4 Denote by f (X1 |X2 , X3 , . . . , Xp ) the conditional density of X1 given X2 , . . . , Xp .

Their proposal has the form (compare (3.69)) B̂c+w = B̂UΛU−1 , (3.72) where Λ is a diagonal shrinkage matrix (the “c+w” stands for “Curds and Whey,” the name they gave to their procedure). Based on optimal prediction in the population setting, they show that Λ has diagonal entries λm = c2m + c2m p N (1 − c2m ) , m = 1, . . . , M, (3.73) where cm is the mth canonical correlation coefficient. Note that as the ratio of the number of input variables to sample size p/N gets small, the shrinkage factors approach 1. Breiman and Friedman (1997) proposed modified versions of Λ based on training data and cross-validation, but the general form is the same. Here the fitted response has the form Ŷc+w = HYSc+w , (3.74) 86 3.

The shrinkage strategy (10.41) tends to eliminate the problem of overfitting, especially for larger data sets. The value of AAE after 800 iterations is 0.31. This can be compared to that of the optimal constant predictor median{yi } which is 0.89. In terms of more familiar quantities, the squared multiple correlation coefficient of this model is R2 = 0.84. Pace and Barry (1997) use a sophisticated spatial autoregression procedure, where prediction for each neighborhood is based on median house values in nearby neighborhoods, using the other predictors as covariates. Experimenting with transformations they achieved R2 = 0.85, predicting log Y .

pages: 295 words: 66,824

A Mathematician Plays the Stock Market by John Allen Paulos

Even a portfolio of stocks from the same sector will be less volatile than the individual stocks in it, while a portfolio consisting of Wal-Mart, Pfizer, General Electric, Exxon, and Citigroup, the biggest stocks in their respective sectors, will provide considerably more protection against volatility. To find the volatility of a portfolio in general, we need what is called the “covariance” (closely related to the correlation coefficient) between any pair of stocks X and Y in the portfolio. The covariance between two stocks is roughly the degree to which they vary together—the degree, that is, to which a change in one is proportional to a change in the other. Note that unlike many other contexts in which the distinction between covariance (or, more familiarly, correlation) and causation is underlined, the market generally doesn’t care much about it.

pages: 217 words: 152

Why Airplanes Crash: Aviation Safety in a Changing World by Clinton V. Oster, John S. Strong, C. Kurt Zorn

the risk that appears to be indicated by the incidents and the appropriate type of accident. Air traffic control operational errors, pilot deviations, and near midair collisions all appear to be indicators of risk of midair collision. Operational errors do not seem closely correlated with midair collisions. Terminal airspace operational errors have essentially no correlation (a correlation coefficient of -0.03) based on the eight years of available data. ARTCC operational errors have been influenced by the introduction of the snitch patch and the controllers' adjustment to it. In the four years of the post-snitch patch era, the correlation with midair collisions is only 0.20. Over the same period, operational errors are actually negatively correlated with the FAA's count of total near midair collisions (-0.84) and critical near midair collisions (-0.83).

Five years of data are simply too little upon which to base a conclusion. The Margin of Safety 119 Near midair collisions are also not strongly correlated with midair collisions. Indeed, near midair collisions as reported to the ASRS show no correlation (-0.08) over the seven years of available data. Midair collisions reported to the FAA have a correlation coefficient of only 0.51, although critical NMACs are somewhat higher at 0.76. The FAA correlation is based on only seven years of data. The poor correlation between potential midair collision incidents and accidents is disappointing to those seeking nonaccident leading indicators of aviation safety, but may not be surprising in light of the characteristics of the incident data.

pages: 681 words: 64,159

Numpy Beginner's Guide - Third Edition by Ivan Idris

The correlaton coefcient takes values between -1 and 1 . The correlaton of a set of values with itself is 1 by defniton. This would be the ideal value; however, we will also be happy with a slightly lower value. Calculate the correlaton coefcient (or, more accurately, the correlaton matrix) with the corrcoef() functon: print("Correlation coefficient", np.corrcoef(bhp_returns, vale_ returns)) The coefcients are as follows: [[ 1. 0.67841747] [ 0.67841747 1. ]] The values on the diagonal are just the correlatons of the BHP and VALE with themselves and are, therefore, equal to 1. In all likelihood, no real calculaton takes place. The other two values are equal to each other since correlaton is symmetrical, meaning that the correlaton of BHP with VALE is equal to the correlaton of VALE with BHP.

For the source code, see the correlation.py fle in this book's code bundle: from __future__ import print_function import numpy as np import matplotlib.pyplot as plt bhp = np.loadtxt('BHP.csv', delimiter=',', usecols=(6,), unpack=True) bhp_returns = np.diff(bhp) / bhp[ : -1] vale = np.loadtxt('VALE.csv', delimiter=',', usecols=(6,), unpack=True) vale_returns = np.diff(vale) / vale[ : -1] covariance = np.cov(bhp_returns, vale_returns) print("Covariance", covariance) print("Covariance diagonal", covariance.diagonal()) print("Covariance trace", covariance.trace()) print(covariance/ (bhp_returns.std() * vale_returns.std())) print("Correlation coefficient", np.corrcoef(bhp_returns, vale_ returns)) difference = bhp - vale avg = np.mean(difference) dev = np.std(difference) print("Out of sync", np.abs(difference[-1] - avg) > 2 * dev) t = np.arange(len(bhp_returns)) plt.plot(t, bhp_returns, lw=1, label='BHP returns') plt.plot(t, vale_returns, '--', lw=2, label='VALE returns') plt.title('Correlating arrays') plt.xlabel('Days') plt.ylabel('Returns') plt.grid() plt.legend(loc='best') plt.show() Q1.

pages: 263 words: 75,455

Quantitative Value: A Practitioner's Guide to Automating Intelligent Investment and Eliminating Behavioral Errors by Wesley R. Gray, Tobias E. Carlisle

Rolling 5-Year Wins Proportion of rolling 5-year periods that a designated strategy beats the identified benchmarks. Rolling 5-Year Wins Proportion of rolling 10-year periods that a designated strategy beats identified benchmarks. Cumulative Drawdown Sum of the rolling 5-year period worst drawdowns for the designated strategy. Correlation Correlation coefficient for a designated strategy and the identified benchmarks, which demonstrates the extent to which a designated strategy and the identified benchmarks move together. RISK AND RETURN Table 12.2 sets out the standard statistical analyses of the Quantitative Value strategy's performance and risk profile, comparing it to the Magic Formula, the Standard & Poor's (S&P) 500 and the MW Index, the market capitalization–weighted index of the universe from which we draw the stocks in the model portfolios.

Rolling 5-Year Wins Proportion of rolling 5-year periods that a designated strategy beats the identified benchmarks. Rolling 10-Year Wins Proportion of rolling 10-year periods that a designated strategy beats identified benchmarks. Cumulative Drawdown Sum of the rolling 5-year period worst drawdowns for the designated strategy. Correlation Correlation coefficient for a designated strategy and the identified benchmarks, which demonstrates the extent to which a designated strategy and the identified benchmarks move together. About the Authors Wesley R. Gray, PhD, is the founder and executive managing member of Empiritrage, LLC, an SEC-Registered Investment Advisor, and Turnkey Analyst, LLC, a firm dedicated to educating and sharing quantitative investment techniques to the general public.

pages: 220 words: 73,451

Democratizing innovation by Eric von Hippel

Controlling for profit expectations, he found that increases in the stickiness of user information were associated with a significant increase in the amount of need-related design undertaken by the user (Kendall correlation coefficient = 0.5784, P < 0.01). Conversely he found that increased stickiness of technology-related information was associated in a significant reduction in the amount of technology design done by the user (Kendall correlation coefficients = 0.4789, P < 0.05). In other words, need-intensive tasks within product-development projects will tend to be done by users, while solutionintensive ones will tend to be done by manufacturers.

pages: 431 words: 132,416

No One Would Listen: A True Financial Thriller by Harry Markopolos

We found out later that several hedge funds believed he was doing this. I created hypothetical baskets using the best-performing stocks and followed his split-strike strategy, selling the call option to generate income and buying the put option for protection. The following week I’d pick another basket. I expected the correlation coefficient—the relationship between Bernie’s returns and the movement of the entire S&P 100—legitimately to be around 50 percent, but it could have been anywhere between 30 percent and 80 percent and I would have accepted it naively. Instead Madoff was coming in at about 6 percent. Six percent! That was impossible.

Pointing to the 6 percent correlation and the 45-degree return line, he said, “That doesn’t look like it came from a finance distribution. We don’t have those kinds of charts in finance.” I was right, he agreed. Madoof’s strategy description claimed his returns were market-driven, yet his correlation coefficient was only 6 percent to the market and his performance line certainly wasn’t coming from the stock market. Volatility is a natural part of the market. It moves up and down—and does it every day. Any graphic representation of the market has to reflect that. Yet Madoff’s 45-degree rise represented a market without that volatility.

pages: 394 words: 85,734

The Global Minotaur by Yanis Varoufakis, Paul Mason

Theocarakis (2011) Modern Political Economics: Making sense of the post-2008 world, London and New York: Routledge. 11. In more technical language, the formulae used to assemble the CDOs assumed that the correlation coefficient between the probability of default across a CDO’s different tranches or slices was constant, small and knowable. 12. Doubt about the constancy of the correlation coefficient (see previous footnote) would have cost them their jobs, particularly as their supervisors did not really understand the formula but were receiving huge bonuses while it was being used. 13. See George Soros (2009) The Crash of 2008 and What It Means: The new paradigm for financial markets, New York: Public Affairs.

Beginning R: The Statistical Programming Language by Mark Gardener

The commands summarized in Table 6-3 enable you to carry out a range of correlation tasks. In the following sections you see a few of these options illustrated, and you can then try some correlations yourself in the activity that follows. Simple Correlation Simple correlations are between two continuous variables and you can use the cor() command to obtain a correlation coefficient like so: > count = c(9, 25, 15, 2, 14, 25, 24, 47) > speed = c(2, 3, 5, 9, 14, 24, 29, 34) > cor(count, speed) [1] 0.7237206 The default for R is to carry out the Pearson product moment, but you can specify other correlations using the method = instruction, like so: > cor(count, speed, method = 'spearman') [1] 0.5269556 This example used the Spearman rho correlation but you can also apply Kendall’s tau by specifying method = “kendall”.

You need to use attach() or with() commands to allow R to “read inside” the data frame and access the variables within. You could also use the \$ syntax so that the command can access the variables as the following example shows: > cor(women\$height, women\$weight) [1] 0.9954948 In this example the cor() command has calculated the Pearson correlation coefficient between the height and weight variables contained in the women data frame. You can also use the cor() command directly on a data frame (or matrix). If you use the data frame women that you just looked at, for example, you get the following: > cor(women) height weight height 1.0000000 0.9954948 weight 0.9954948 1.0000000 Now you have a correlation matrix that shows you all combinations of the variables in the data frame.

In regression you are taking the analysis further and assuming a mathematical, and therefore predictable, relationship between the variables. The results of regression analysis show the slope and intercept values that describe this relationship. The R squared value that you obtain from the regression is the square of the correlation coefficient from the Pearson correlation, which demonstrates the similarities between the methods. The result shows you the coefficients for the regression, that is, the intercept and the slope. To see more details you should save your regression as a named object; then you can use the summary() command like so: > fw.lm = lm(count ~ speed, data = fw) > summary(fw.lm) Call: lm(formula = count ~ speed, data = fw) Residuals: Min 1Q Median 3Q Max -13.377 -5.801 -1.542 5.051 14.371 Coefficients: Estimate Std.

pages: 459 words: 144,009

Upheaval: Turning Points for Nations in Crisis by Jared Diamond

One method by which social scientists have tested this belief is to compare, among different countries, the correlation coefficients between incomes (or income ranks within people of their generation) of adults and the incomes of their parents. A correlation coefficient of 1.0 would mean that relative incomes of parents and of their adult children are perfectly correlated: all high-income people are children of high-income parents, all low-income people are children of low-income parents, kids from low-income families have zero chance of achieving high incomes, and socio-economic mobility is zero. At the opposite extreme, if the correlation coefficient were zero, it would mean that children of low-income parents have as good a chance of achieving high incomes as do children of high-income parents, and socio-economic mobility is high.

pages: 268 words: 89,761

Unhealthy societies: the afflictions of inequality by Richard G. Wilkinson

If increases in GNPpc over time were simply understated, it might be thought that this would not mask a statistical relationship between health and GNPpc: the extent to which societies benefited from qualitative changes in output would be a constant function of their growth rates. If this were so, then understated growth would change the units rather than weaken the correlation between the two. It would tend to make any given increase in income appear more health effective. In technical terms: rather than weakening the correlation coefficient it would increase the size of the regression coefficient. However, it could be argued that the spread of better products does not depend simply on the expenditure which results from the few per cent of income growth. Much nearer to the truth is that, as earlier forms of goods are made obsolete and replaced in the shops by new models and lines, the whole flow of expenditure is applied to the current range of goods, including new goods and ones in which the quality has changed.

Since then van Doorslaer et al. have reported that differences in self-reported illness were greatest in countries whose income differences were greatest (van Doorslaer Income distribution and health 89 et al. 1996). The relationship between measures of inequality in income and in illness was very close: across the USA and the eight European countries for which they had data, the correlation coefficient was 0.87. The methods used in each study were quite different. Kunst and Mackenbach classified people according to occupation in one of their studies and by education in another, and they concluded that occupational and educational differences in mortality were greater in countries where income differences were greater.

pages: 353 words: 88,376

The Investopedia Guide to Wall Speak: The Terms You Need to Know to Talk Like Cramer, Think Like Soros, and Buy Like Buffett by Jack (edited By) Guinan

Related Terms: • Bond • Municipal Bond • Yield to maturity—YTM • Debt Financing • Yield Correlation What Does Correlation Mean? In the investment world, correlation is a statistical measure of how two securities move in relation to each other. Correlations are used in advanced portfolio management. Investopedia explains Correlation Correlation is expressed as the correlation coefficient, which ranges between –1 and +1. Perfect positive correlation (a correlation coefficient of +1) means that as one security moves up or down, the other security will move lockstep in the same direction. Perfect negative correlation means that when one security moves in one direction, the other security will move by an equal amount in the opposite direction.

pages: 287 words: 44,739

Guide to business modelling by John Tennent, Graham Friend, Economist Group

The coefficient of determination R2 (which is calculated automatically by most spreadsheet packages) indicates how much of the variation in Y is explained by the explanatory variables. The greater the value of R2, the more the variation in the dependent variable is explained by the selected independent variables. The square root of the coefficient of determination is the product moment correlation coefficient in the case of linear regression of a straight line. The product moment correlation is a number between 1 and ⫺1. If r ⫽ 1 then there is a perfect, positive relationship between the dependent and explanatory variable. A perfect relation implies that every data point lies on a straight line.

Chart 10.14 Regression equation for monthly gross connections against time The R2 value is very low at 0.315. This implies that only 31.5% of the variation in gross connections is explained by time, so any forecast based on this regression equation will be liable to considerable error. The correlation coefficient is the square root of R2⫽SQR(0.315)⫽0.561. Although this value is low it is possible to show, using significance testing, that time is still a significant determinant of gross connections. Regression techniques 99 This procedure is quick and simple to use. However, to develop a forecast it is necessary to use the equation of the straight line.

The Rise and Decline of Nations: Economic Growth, Stagflation, and Social Rigidities by Mancur Olson

Kwang Choi, "A Study of Comparative Rates of Economic Growth" (forthcoming, Iowa State University Press) and Kwang Choi, "A Statistical Test of the Political Economy of Comparative Growth Rates Model," in Mueller, The Political Economy of Growth. 34. Spearman rank correlation coefficients between years since statehood and LPI, PN, and per capita LP/, PN were respectively -.52, -.67, -.52, and -.52, and the correlation coefficients were in every case significant. 35. Farm organization membership need not be correlated with union membership, but farm groups focus almost exclusively on the farm policies of the federal government, and any losses in output due to them must fall mainly on consumers throughout the United States, rather than in the state in which the farmers are organized, so farm organization membership probably should not be included in tests on the forty-eight contiguous states.

pages: 416 words: 106,532

Cryptoassets: The Innovative Investor's Guide to Bitcoin and Beyond: The Innovative Investor's Guide to Bitcoin and Beyond by Chris Burniske, Jack Tatar

Cryptoassets have near-zero correlation to other capital market assets. The best explanation for this is that cryptoassets are so new that many capital market investors don’t play in the same asset pools. Therefore, cryptoassets aren’t dancing to the same rhythm of information as traditional capital market assets, at least not yet. Figure 7.19 The correlation coefficient and effects of diversification on risk Source: A Random Walk Down Wall Street, Burton G. Malkiel, 2015 Figure 7.19 clearly shows that if an asset is zero correlated to other assets in a portfolio, then “considerable risk reduction is possible.” In quantitative terms, reducing risk can be seen by a decrease in the volatility of the portfolio.

See Bitcoin Tracker One Cold storage, 221–222 Collaboration, 111 community and, 56 platforms for, 159 Collateralized mortgage obligations (CMOs), 4–5 Colored coins, 53 Commodities, 80, 172, 276–277 Commodities Futures Trading Commission (CFTC), 107, 112, 224, 276 Communication, 14 Communication Nets, xxiii Community, 57, 62 collaboration and, 56 of computers, 18 developers and, 182 Companies, 28, 63, 118 as incumbents, 264–273 interface services by, 113 OTC by, 216 as peer-to-peer, 13 perspective of, 249–250 risk and, 75 support and, 198–200 technology and, 264–265 value of, 152 venture capitalism for, 248 Competition, 16, 214 Compound annual growth rate (CAGR), 118–119 Compound annual returns, 87, 88, 103–104 Computer scientists, 60 Computers blockchain technology and, 26, 186 community of, 18 as miners, 16 for mining, 212 private keys on, 226 supercomputers as, 59 Consortium, 272–273 Consumable/Transformable (C/T) Assets, 109–110 Content, 174 Corbin, Abel, 164–165 Cornering, 163–166 cryptoassets and, 166–168 Correlation coefficient, 101 Correlation of returns, 74–76 Correlations, 122 assets and, 74 Bitcoin and, 133 cryptoassets and, 101–102 market behavior and, 132–135 Counterparty, 53–54 CPUs. See Central processing units Credit, 153 assets and, 143 issuers quality of, 239 Credulity, 141 The Crowd: A Study of the Popular Mind (Le Bon), 140 Crowd theory, 141 Crowdfunding, 60 Internet and, 250–254, 256 for investors, 250–252 for projects, 254 regulations and, 250 Crowds, 137–153 Crowdsale, 257 Cryptoassets.

pages: 397 words: 109,631

Mindware: Tools for Smart Thinking by Richard E. Nisbett

A correlation of .8 corresponds to the degree of association you find between scores on the math portion of the Scholastic Aptitude Test (SAT) at one testing and scores on that test a year later—quite high but still plenty of room for difference between the two scores on average. Correlation Does Not Establish Causality Correlation coefficients are one step in assessing causal relations. If there is no correlation between variable A and variable B, there (probably) is no causal relation between A and B. (An exception would be when there is a third variable C that masks the correlation between A and B when there is in fact a causal relation between A and B.)

Coding Is the Key to Thinking Statistically I’m going to ask you some questions concerning your beliefs about what you think the correlation is between a number of pairs of variables. The way I’ll do that is to ask you how likely it is that A would be greater than B on one occasion given that A was greater than B on another occasion. Your answers in probability terms can be converted to correlation coefficients by a mathematical formula. Note that if you say “50 percent” for a question below, you’re saying that you think there’s no relationship between behavior on one occasion and behavior on another. If you say “90 percent,” you’re saying that there is an extremely strong relationship between behavior on one occasion and behavior on another.

pages: 339 words: 112,979

Unweaving the Rainbow by Richard Dawkins

Even if their verdicts are wrong, you'd think their methods would be systematic enough at least to agree in producing the same wrong verdicts! Alas, as shown in a study by G. Dean and colleagues, they don't even achieve this minimal and easy benchmark. For comparison, when different assessors judged people on their performance in structured interviews, the correlation coefficient was greater than 0.8 (a correlation coefficient of 1.0 would represent perfect agreement, –1.0 would represent perfect disagreement, 0.0 would represent complete randomness or lack of association; 0.8 is pretty good). Against this, in the same study, the reliability coefficient for astrology was a pitiable 0.1, comparable to the figure for palmistry (0.11), and indicating near total randomness.

pages: 743 words: 189,512

The Big Fat Surprise: Why Butter, Meat and Cheese Belong in a Healthy Diet by Nina Teicholz

In 1999, when the Seven Countries study’s lead Italian researcher, Alessandro Menotti, went back twenty-five years later and looked at data from the study’s 12,770 subjects, he noticed an interesting fact: the category of foods that best correlated with coronary mortality was sweets. By “sweets,” he meant sugar products and pastries, which had a correlation coefficient with coronary mortality of 0.821 (a perfect correlation is 1.0). Possibly this number would have been higher had Menotti included chocolate, ice cream, and soft drinks in his “sweets” category, but those fell under a different category and, he explained, would have been “too troublesome” to recode. By contrast, “animal food” (butter, meat, eggs, margarine, lard, milk, and cheese) had a correlation coefficient of 0.798, and this number likely would have been lower had Menotti excluded margarine.

Public Health Nutrition 8, no. 6 (2005): 666. “we should not” . . . “the ideal thing all the time”: Daan Kromhout, interview with author, October 4, 2007. he knew it would go unnoticed: Keys, Aravanis, and Sdrin, “Diets of Middle-Aged Men in Two Rural Areas of Greece,” 577. category of foods . . . which had a correlation coefficient: Alessandro Menotti et al., “Food Intake Patterns and 25-Year Mortality from Coronary Heart Disease: Cross-Cultural Correlations in the Seven Countries Study,” European Journal of Epidemiology 15, no. 6 (1999): 507–515. “too troublesome” to recode: Alessandro Menotti, interview with author, July 24, 2008.

pages: 533 words: 125,495

Rationality: What It Is, Why It Seems Scarce, Why It Matters by Steven Pinker

The residuals also allow us to quantify how correlated the two variables are: the shorter the bands, as a proportion of how splayed out the entire cluster is left to right and up and down, the closer the dots are to the line, and the higher the correlation. With a bit of algebra this can be converted into a number, r, the correlation coefficient, which ranges from –1 (not shown), where the dots fall in lockstep along a diagonal from northwest to southeast; through a range of negative values where they splatter diagonally along that axis; through 0, when they are an uncorrelated swarm of gnats; through positive values where they splatter southwest to northeast; to 1, where they lie perfectly along the diagonal.

See moral progress —recommendations to strengthen accountability for lying and disinformation, 313, 314, 316–17 avoidance of sectarian symbolism, 312 educational curricula, 314–15 evidence-based evaluation, 312, 317 incentive structures, 315–17 in journalism, 314, 316, 317 norms valorizing, 311–13, 315 in punditry, 317 “Republican party of stupid,” 312–13, 357n73 scientists in legislatures, 312 in social media, 313, 316–17 viewpoint diversity in higher ed, 313–14 Rationality Community, 149–50, 312 Rationality Quotient, 311 Rawls, John, 69 realism, universal, 300–301 reality, motivating rationality, 41–42, 288, 298, 309, 320 reality mindset Active Open-Mindedness/Openness to Evidence and, 310–11, 324, 356–57n67 definition, 299–300 mythology mindset, border with, 303 as unnatural, 300–301 reason, in definition of “rationality,” 36 Rebel Without a Cause, 59, 236, 344n29 reciprocity, norms of, 5 recursion, 71, 108 reflectiveness Cognitive Reflection Test, 8–11, 50 definition, 311 intelligence correlating with, 311 openness to evidence correlating with, 311 and reasoning competence, 323, 324 resistance to cognitive illusions and, 311 unreflective thinking, 8–10, 311 of weird beliefs, 299 regression correlation coefficient (r), 250–51 definition, 248, 252 equation for, 272, 278–81 general linear model, 272 human vs., accuracy of, 278–80 instrumental variable regression, 267–68 multiple regression, 270–72, 271 regression line, 248–49, 253–54, 270 residuals, 249–50, 270–71 regression discontinuity, 266–67 regression line, 248–49, 253–54, 270 regression to the mean and bell curve distribution, 253 definition, 252–53 imperfect correlation as producing, 254 regression line, 253–54 scatterplots and, 253–54 as statistical phenomenon, 253 unawareness of, 254–56, 320, 353n13 regret avoidance, 17, 190 relativism and relativists argument against rationality, 39–40 as hypocritical, 42 morality and, 42, 66–67 religion argument from authority, 90 the cluster illusion and, 147 forbidden base rates and, 163–66 the Golden Rule in, 68–69 heretical counterfactuals, 64–65 monotheism, 40 the mythology mindset and, 301–2, 307 persecution of, progress against, 330–31 See also God Rendezvous game, 233–34 replicability crisis in science Bayesian reasoning failures and, 159–61 preregistration as remedy, 145–46 questionable research practices and, 145–46, 160, 353n13 science journalism and, 161–62 statistical significance and, 225 Texas sharpshooter fallacy, 144–46, 160 Winner’s Curse, 256 representativeness heuristic, 27, 155–56 Republican Party and Republicans calling Democrats socialists, 83–84 expressive rationality and, 298 Fox News and, 267, 268, 296 pizza classified as vegetable by, 101 politically motivated numeracy, 292–94 rehabilitating, 312–13, 357n73 See also left and right (political); politics reputation, 47, 237, 242, 308, 313 resistance to evidence.

pages: 823 words: 220,581

Debunking Economics - Revised, Expanded and Integrated Edition: The Naked Emperor Dethroned? by Steve Keen

Though the huge fiscal and monetary stimulus packages also played a role, changes in debt-financed demand dominate economic performance. One statistical indicator of the importance of debt dynamics in causing both the Great Depression and the Great Recession and the booms that preceded them is the correlation coefficient between changes in debt and the level of unemployment. Over the whole period from 1921 till 1940, the correlation coefficient was minus 0.83, while over the period from 1990 till 2011, it was minus 0.91 (versus the maximum value it could have taken of minus one). A correlation of that scale, over time periods of that length, when economic circumstances varied from bust to boom and back again, is staggering. 13.31 Debt-financed demand and unemployment, 1990–2011 The Credit Impulse confirms the dominant role of private debt.

In Sharpe’s words: In order to derive conditions for equilibrium in the capital market we invoke two assumptions. First, we assume a common pure rate of interest, with all investors able to borrow or lend funds on equal terms. Second, we assume homogeneity of investor expectations: investors are assumed to agree on the prospects of various investments – the expected values, standard deviations and correlation coefficients described in Part II. Needless to say, these are highly restrictive and undoubtedly unrealistic assumptions. However, since the proper test of a theory is not the realism of its assumptions but the acceptability of its implications, and since these assumptions imply equilibrium conditions which form a major part of classical financial doctrine, it is far from clear that this formulation should be rejected – especially in view of the dearth of alternative models leading to similar results.

The Age of Turbulence: Adventures in a New World (Hardback) - Common by Alan Greenspan

THE M O D E S OF C A P I T A L I S M at the same time, Germany ranks among the highest in terms of the freedom of its people to open and close businesses, property-rights protection, and the overall rule of law. France (number forty-five) and Italy (number sixty) have profiles that are similarly mixed. The ultimate test of the usefulness of such a scoring process is whether it correlates with economic performance. And it does. The correlation coefficient of 157 countries between their "Economic Freedom Score" and the log of their per capita incomes is 0.65, impressive for such a motley body of data.* Thus, we are left with a critical question: Granted that open competitive markets foster economic growth, is there an optimum trade-off between economic performance and the competitive stress it imposes on the one hand, and the civility that, for example, the continental Europeans and many others espouse?

Accordingly the weighted correlation between national saving rates and domestic investment rates for countries or regions representing virtually all of the world's gross domestic product, a measure of the degree of home bias, declined from a coefficient of around 0.95 in 1992, where it had hovered since 1970, to an estimated 0.74 in 2005. (If in every country saving equaled investment—that is, if there were 100 percent home bias—the correlation coefficient would be 1.0. On the other hand, if there were no home bias, and the amount of domestic saving bore no relationship to the amount and location of investments, the coefficient would be 0.)* Only in the past decade has expanding trade been associated with the emergence of ever-larger U.S. trade and current account deficits, matched by a corresponding widening of the aggregate external surpluses of many of our trading partners, most recently including China.

Although foreign investors *The persistent divergence subsequent to t h e creation of t h e euro of m a n y prices of identical goods a m o n g m e m b e r countries of t h e euro area is analyzed in John H. Rogers (2002). For t h e case of U.S. and Canadian prices, see Charles Engel and John H. Rogers ( 1 9 9 6 ) . t T h e correlation coefficient measures of h o m e bias have flattened o u t since 2 0 0 0 . So have t h e measures of dispersion. This is consistent w i t h t h e United States' accounting for a rising share of deficits. 361 More ebooks visit: http://www.ccebook.cn ccebook-orginal english ebooks This file was collected by ccebook.cn form the internet, the author keeps the copyright.

Mastering Machine Learning With Scikit-Learn by Gavin Hackeling

An r-squared score of one indicates that the response variable can be predicted without any error using the model. An r-squared score of one half indicates that half of the variance in the response variable can be predicted using the model. There are several methods to calculate r-squared. In the case of simple linear regression, r-squared is equal to the square of the Pearson product moment correlation coefficient, or Pearson's r. [ 29 ] www.it-ebooks.info Linear Regression Using this method, r-squared must be a positive number between zero and one. This method is intuitive; if r-squared describes the proportion of variance in the response variable explained by the model, it cannot be greater than one or less than zero.

pages: 433 words: 53,078

Be Your Own Financial Adviser: The Comprehensive Guide to Wealth and Financial Planning by Jonquil Lowe

For example, equities and property tend to do well in times of inflation, unlike cash and bonds. But cash and bonds may produce solid returns in an economic downturn, while property and equities tend to suffer. The extent to which different investments or assets are correlated can be measured and represented by a statistic called a ‘correlation coefficient’. A coefficient of 1 would mean that two asset classes moved in exactly the same way (so there would not be any point combining the assets). A coefficient of zero would mean the asset classes were completely uncorrelated. Most coefficients lie between these two extremes. A negative coefficient means that positive performance for one asset class is associated with negative performance from the other.

A negative coefficient means that positive performance for one asset class is associated with negative performance from the other. (A coefficient of –1 would mean the assets were perfectly negatively correlated with the risk of losses on one asset being completely offset by the chance of gains on the other, so eliminating risk altogether for a portfolio made up of these two assets.) Table 10.1 on p. 301 shows the correlation coefficient for different pairs of asset class using data for the period from 1997 to 2009. The ideal for investors is to combine asset classes that have low or negative correlations, say, less than 0.5. The global financial crisis that hit the world in 2007 demonstrated that, in some situations, the relative independence of the asset classes can break down, with all delivering poor performance simultaneously.

Trend Commandments: Trading for Exceptional Returns by Michael W. Covel

Consider this nearly 40-year track record of trend trading wealth building: Chart 1: Bill Dunn Unit Value Log Scale DUNN Composite Performance: \$500,000 \$570,490 10 Drawdowns Greater Than -25% October 1974 through January 2011 Average Major Drawdown: 38% \$100,000 -40% Compound Annual Rate of Return -27% -60% -34% \$62,375 DUNN Composite: 19.09% S&P 500 (Total return) : 11.85% -29% -34% -45% -35% -51% -43% \$10,000 -30% Correlation Coefficient -0.05 Past Performance is Not Necessarily Indicative of Future Results -28% -45% Includes Notional and Proprietary Funds All Net of Pro Forma Fees and Expenses -52% ' 70 ' 71 ' 72 ' 73 ' 74 ' 75 ' 76 ' 77 ' 78 ' 79 ' 80 ' 81 ' 82 ' 83 ' 84 ' 85 ' 86 ' 87 ' 88 ' 89 ' 90 ' 91 ' 92 ' 93 ' 94 ' 95 ' 96 ' 97 ' 98 ' 99 ' 00 ' 01 ' 02 ' 03 ' 04 ' 05 ' 06 ' 07 ' 08 ' 09 ' 10 ' 11 \$1,000 That picture is worth a thousand words.

pages: 204 words: 58,565

Keeping Up With the Quants: Your Guide to Understanding and Using Analytics by Thomas H. Davenport, Jinho Kim

., records in a database) into groups (called clusters) so that objects within clusters are similar in some manner while objects across clusters are dissimilar to each other. Clustering is a main task of exploratory data mining, and a common technique for statistical data analysis used in many fields. Correlation: The extent to which two or more variables are related to one another. The degree of relatedness is expressed as a correlation coefficient, which ranges from −1.0 to +1.0. Correlation = +1 (Perfect positive correlation, meaning that both variables always move in the same direction together) Correlation = 0 (No relationship between the variables) Correlation = −1 (Perfect negative correlation, meaning that as one variable goes up, the other always trends downward) Correlation does not imply causation.

pages: 935 words: 267,358

Capital in the Twenty-First Century by Thomas Piketty

But this is a different issue from skill and earned income mobility, which is what is of interest here and is the focal point of these measurements of intergenerational mobility. The data used in these works do not allow us to isolate mobility of capital income. 28. The correlation coefficient ranges from 0.2–0.3 in Sweden and Finland to 0.5–0.6 in the United States. Britain (0.4–0.5) is closer to the United States but not so far from Germany or France (0.4). Concerning international comparisons of intergenerational correlation coefficients of earned income (which are also confirmed by twin studies), see the work of Markus Jantti. See the online technical appendix. 29. The cost of an undergraduate year at Harvard in 2012–2013 was \$54,000, including room and board and various other fees (tuition in the strict sense was \$38,000).

According to the available data, the answer seems to be no: the intergenerational correlation of education and earned incomes, which measures the reproduction of the skill hierarchy over time, shows no trend toward greater mobility over the long run, and in recent years mobility may even have decreased.26 Note, however, that it is much more difficult to measure mobility across generations than it is to measure inequality at a given point in time, and the sources available for estimating the historical evolution of mobility are highly imperfect.27 The most firmly established result in this area of research is that intergenerational reproduction is lowest in the Nordic countries and highest in the United States (with a correlation coefficient two-thirds higher than in Sweden). France, Germany, and Britain occupy a middle ground, less mobile than northern Europe but more mobile than the United States.28 These findings stand in sharp contrast to the belief in “American exceptionalism” that once dominated US sociology, according to which social mobility in the United States was exceptionally high compared with the class-bound societies of Europe.

Once the American Dream: Inner-Ring Suburbs of the Metropolitan United States by Bernadette Hanlon

Table A.7 174 / Appendix TABLE A.7 RESULTS OF PEARSON CORRELATION BETWEEN INDEX SCORE AND CHANGE IN THE MEDIAN HOUSEHOLD INCOME RATIO FROM 1980 TO 2000 Variables Index score Change in median household income ratio from 1980 to 2000 Change in median household income ratio from 1980 to 2000 1 −0.801a −0.801a 1 3,428 3,428 N a Correlation Index score is significant at the 0.01 level (2-tailed). shows a Pearson’s Correlation between these two variables of −0.801. Pearson’s Correlation is a measure of correlation between two variables— that is, a measure of the tendency of variables to increase or decrease together. The correlation coefficient of −0.801 indicates that 80 percent of the variance in income is explained by variance in index score. The index score and the change in median household income ratio are highly negatively correlated. As the index score increases, the median household income ratio increases less over time.

pages: 270 words: 73,485

Hubris: Why Economists Failed to Predict the Crisis and How to Avoid the Next One by Meghnad Desai

This is to allow for the basic uncertainty of all economic events, as well as to allow for many other variables which have to be omitted to keep the relationship simple. If the basic equation is sound, then it will explain a large part of the variation in the variable we are interested in, in our case y, the amount bought of a commodity. A measure of the “goodness of fit” is the correlation coefficient r or its square R2 (R squared). Many equations together constitute a model and there are more sophisticated measures of the explanatory powers of a model. The use of econometric techniques is widespread now in public and private sector decision-making. Increasingly numbers have become an indispensable part of the toolkit of economists.

pages: 306 words: 78,893

After the New Economy: The Binge . . . And the Hangover That Won't Go Away by Doug Henwood

Hours worked was estimated by multiplying the BL5 figure for total employment by average weekly hours in each of the component industries, adding them together, and subtracting the result from a similar estimate of total manufacturing hours worked. While not exact, the approximation is pretty good; an estimate of total manufacturing productivity using this technique had a correlation coefficient of .86 with the official index. See text for discussion. the way we live and work, sometimes to the good, sometimes not. Do they make a 28% annual contribution to the growth of human happiness? Closely related to the productivity argument is a claim about innovation: that we Hve in a time of new product development without any his- torical precedent.

pages: 209 words: 13,138

Empirical Market Microstructure: The Institutions, Economics and Econometrics of Securities Trading by Joel Hasbrouck

This conforms to the usual intuition that positive correlation aggravates total portfolio risk. On the other hand, if we assume (as before) that the dealer is starting at his optimum, then B1 = P1 − ασ12 /2. Surprisingly, this is the same result as in the one-security case. In particular, the correlation coefficient drops out. This is a consequence of offsetting effects. The optimal n1 and n2 in equation (11.3) depends negatively on ρ, leaving the bracketed term invariant to changes in ρ. (Although this offset is a general feature of the problem, the complete disappearance of ρ in the final expression for the bid is a consequence of CARA utility.) 11.3 Empirical Analysis of Dealer Inventories 11.3.1 A First Look at the Data Changes in the dealer’s position reveal the dealer’s trades, which may disclose strategy and profitability.

pages: 325 words: 73,035

Who's Your City?: How the Creative Economy Is Making Where to Live the Most Important Decision of Your Life by Richard Florida

Seligman, “Beyond Money: Toward an Economy of Well-Being,” Psychological Science in the Public Interest 5, 1, 2004, pp. 1-31.Betsy Stevenson and Justin Wolfers, “Economic Growth and Subjective Wellbeing: reassessing the Easterlin Paradox,” Wharton School, University of Pennsylvania, May 9, 2008, http://bpp.wharton.upenn.edu/jwolfers/Papers/EasterlinParadox.pdf. 3 Also see Angus Deaton, “Income, Aging, Health, and Wellbeing Around the World: Evidence from the Gallup World Poll,” Center for Health and Wellbeing, Research Program in Development Studies, Princeton University, August 2007. 4 Nick Paumgarten, “There and Back Again,” New Yorker, April 16, 2007. 5 Robert Manchin, “The Emotional Capital and Desirability of European Cities,” Gallup Europe, presented at the European Week of Cities and Regions, Brussels, October 2007. 6 The correlation coefficients between overall happiness and various factors are as follows: financial satisfaction (.369), job satisfaction (.367), place satisfaction (.303). Compare with income (.153), homeownership (.126), and age (.06). The regression coefficients (from an ordered probit regression) are as follows: financial satisfaction (.342), place satisfaction (.254), job satisfaction (.254).

pages: 249 words: 77,342

The Behavioral Investor by Daniel Crosby

The authors of ‘Positive Illusions and Forecasting Errors in Mutual Fund Investment Decisions’ discovered that most participants had consistently overestimated both the future and past performance of their investments.70 One-third of those who believed that they had outperformed the market had actually lagged by at least 5% and another quarter of people lagged by 15% or greater. Even more damning research is found by Glaser and Weber who discovered that, “Investors are unable to give a correct estimate of their own past portfolio performance. The correlation coefficient between return estimates and realized returns was not distinguishable from zero.”71 The finding that investors would misstate their returns is not entirely surprising, but the size and scope of the problem is. Only 30% of those surveyed considered themselves to be “average” investors and the average overestimation of returns was 11.5% per year!

pages: 266 words: 76,299

Ever Since Darwin: Reflections in Natural History by Stephen Jay Gould

He has found an astonishing number of inconsistencies and downright inaccuracies. For example, the late Sir Cyril Burt, who generated the largest body of data on identical twins reared apart, pursued his studies of intelligence for more than forty years. Although he increased his sample sizes in a variety of “improved” versions, some of his correlation coefficients remain unchanged to the third decimal place—a statistically impossible situation.5 IQ depends in part upon sex and age; and other studies did not standardize properly for them. An improper correction may produce higher values between twins not because they hold genes for intelligence in common, but simply because they share the same sex and age.

pages: 701 words: 199,010

The Crisis of Crowding: Quant Copycats, Ugly Models, and the New Crash Normal by Ludwig B. Chincarini

Unfortunately, it is impossible to have an average correlation of −1 with more than two strategies in a portfolio. As you add more strategies, however, the conditions necessary for a very low-risk portfolio grow less stringent. Consider a portfolio of n trading strategies. Assuming a similar variance for each strategy, , and that any pair of strategies has the same correlation coefficient, . If all strategies are equally weighted (that is, ) and individual strategy returns are positive, the portfolio variance is: (A.4) Construct a portfolio with a large number of positions that have an average correlation of zero, and the portfolio risk decreases toward zero. For the purposes of this analysis, assume that this was LTCM’s driving concept.

., k = 1.96 for a 97.5% confidence interval).9 Table A.1 presents the potential VaR calculations at a 99% confidence level for a normal distribution (k = 2.33) and a capital base of \$4.8B (the amount that LTCM had at the beginning of 1998). The VaR numbers are presented as monthly numbers. Given the correlation coefficient, this represents what might have been expected to occur in any given month at LTCM. TABLE A.1 Sensitivity of VaR to Strategy Correlations Table A.1 shows that an unlevered fund’s standard deviation was 0.0951% per month and 0.6723% per month with a correlation of 0 and 1 respectively.

pages: 297 words: 84,009

Big Business: A Love Letter to an American Anti-Hero by Tyler Cowen

Here too we should be cautious about how grand a conclusion we draw from a single study, but this is suggestive evidence that the workplace often serves a significant protective and equalizing function when it comes to personal stress. Furthermore, the Kahneman and Krueger research generates a broadly similar result. The positive affect associated with the workday is not closely related to the features we usually associate with a “good” job. (For instance, the correlation coefficient of positive affect in the workplace with “excellent benefits” is only about 0.10.) People with lower-quality jobs still get a lot of the benefits from the positive affect associated with work. Here’s a simple and probably familiar story from Elizabeth Bernstein, writing in the Wall Street Journal.

pages: 345 words: 87,745

The Power of Passive Investing: More Wealth With Less Work by Richard A. Ferri

closed-end fund A mutual fund that has a fixed number of shares, usually listed on a major stock exchange. commodities Unprocessed goods, such as grains, metals, and minerals, traded in large amounts on a commodities exchange. consumer price index (CPI) A measure of the price change in consumer goods and services. The CPI is used to track the pace of inflation. correlation coefficient A number between −1 and 1 that measures the degree to which two variables are linearly related. cost basis The original cost of an investment. For tax purposes, the cost basis is subtracted from the sale price to determine any capital gain or loss. country risk The possibility that political events (e.g., a war, national elections); financial problems (e.g., rising inflation, government default); or natural disasters (e.g., an earthquake, a poor harvest) will weaken a country’s economy and cause investments in that country to decline.

pages: 901 words: 234,905

The Blank Slate: The Modern Denial of Human Nature by Steven Pinker

Chapter 16 Politics I often think it’s comical How nature always does contrive That every boy and every gal, That’s born into the world alive, Is either a little Liberal, Or else a little Conservative!1 GILBERT AND SULLIVAN got it mostly right in 1882: liberal and conservative political attitudes are largely, though far from completely, heritable. When identical twins who were separated at birth are tested in adulthood, their political attitudes turn out to be similar, with a correlation coefficient of. 62 (on a scale from-1 to +1).2 Liberal and conservative attitudes are heritable not, of course, because attitudes are synthesized directly from DNA but because they come naturally to people with different temperaments. Conservatives, for example, tend to be more authoritarian, conscientious, traditional, and rule-bound.

It is mathematically meaningful to say that a certain percentage of the variance in a group overlaps with one factor (perhaps, though not necessarily, its cause), another percentage overlaps with a second factor, and so on, the percentages adding up to 100. The degree of overlap may be measured as a correlation coefficient, a number between-1 and +1 that captures the degree to which people who are high on one measurement are also high on another measurement. It is used in behavioral genetic research as an estimate of the proportion of variance accounted for by some factor.3 Heritability is the proportion of variance in a trait that correlates with genetic differences.

pages: 317 words: 106,130

The New Science of Asset Allocation: Risk Management in a Multi-Asset World by Thomas Schneeweis, Garry B. Crowder, Hossein Kazemi

The break-even (Rc) and excess break-even rate of return (EBK) is often computed as follows: ⎛ E (Rp ) − Rf ⎞ E ( Rc ) = ⎜ ⎟⎠ ( ρcp ) σ c + Rf ⎝ σp ⎤ ⎡⎛ E (Rp ) − Rf ⎞ EBK = Rc − ⎢⎜ ⎟⎠ ( ρcp ) σ c + Rf ⎥ ⎝ σ p ⎦ ⎣ where E(Rc) = Break-even rate of return required for the asset to improve the Sharpe Ratio of alternative index p Rc = Rate of return on asset c Rf = Riskless rate of return E(Rp) = Rate of return on alternative index p ρcp = Correlation coefficient between asset c and alternative benchmark p σc = Standard deviation of asset c σp = Standard deviation of alternative index p First, it is important to realize that the above expression is based on the assumption that only mean and variance matter in evaluating the risk-return profile of a portfolio.

pages: 571 words: 105,054

Tamhane, and F. Bretz (2010): Multiple Testing Problems in Pharmaceutical Statistics, 1st ed. CRC Press. Dudoit, S. and M.J. van der Laan (2008): Multiple Testing Procedures with Applications to Genomics, 1st ed. Springer. Fisher, R.A. (1915): “Frequency distribution of the values of the correlation coefficient in samples of an indefinitely large population.” Biometrika (Biometrika Trust), Vol. 10, No. 4, pp. 507–521. Hand, D. J. (2014): The Improbability Principle, 1st ed. Scientific American/Farrar, Straus and Giroux. Harvey, C., Y. Liu, and H. Zhu (2013): “. . . And the cross-section of expected returns.”

pages: 417 words: 103,458

The Intelligence Trap: Revolutionise Your Thinking and Make Wiser Decisions by David Robson

Stanovich has now spent more than two decades building on those foundations with a series of carefully controlled experiments. To understand his results, we need some basic statistical theory. In psychology and other sciences, the relationship between two variables is usually expressed as a correlation coefficient between 0 and 1. A perfect correlation would have a value of 1 – the two parameters would essentially be measuring the same thing; this is unrealistic for most studies of human health and behaviour (which are determined by so many variables), but many scientists would consider a ‘moderate’ correlation to lie between 0.4 and 0.59.11 Using these measures, Stanovich found that the relationships between rationality and intelligence were generally very weak.

pages: 389 words: 109,207

Fortune's Formula: The Untold Story of the Scientific Betting System That Beat the Casinos and Wall Street by William Poundstone

For reasons mathematical, psychological, and sociological, it is a good idea to use a money management system that is relatively forgiving of estimation errors. Fat Tails and Leverage Suppose you’re betting on a simultaneous toss of coins believed to have a 55 percent chance of coming up heads, as depicted on the previous page. But on this toss, only 45 percent of the coins are heads. Call it a “fat tail” event, or a failure of correlation coefficients, or a big dumb mistake in somebody’s computer model. What then? The Kelly bettor cannot be ruined in a single toss. (He is prepared to survive the worst-case scenario, of zero heads.) In this situation, with many coins, the Kelly bettor will stake just short of his full bankroll. He wins only 45 percent of the wagers, doubling the amount bet on each coin that comes up heads.

pages: 370 words: 107,983

Rage Inside the Machine: The Prejudice of Algorithms, and How to Stop the Internet Making Bigots of Us All by Robert Elliott Smith

Processing the lab’s big data required statistical mathematics, so in 1911 Pearson (who already held a chair in Applied Mathematics at UCL) merged the biometric and eugenics laboratories to form the Department of Applied Statistics, the first university statistics department in the world. Pearson went on to create the Pearson correlation coefficient, one of the most fundamental calculations in statistics. In fact, his work is so foundational to statistics that he was offered a knighthood (which he declined based on his personal commitment to socialism). The UCL building which once housed the Department of Statistics bears his name. Pearson also founded The Annals of Eugenics journal (which now exists as the prominent Annals of Genetics), the masthead of which originally included the famous (mis)quote from Charles Darwin.

pages: 421 words: 110,272

Deaths of Despair and the Future of Capitalism by Anne Case, Angus Deaton

National Research Council, 2005, “Firearms and suicide,” in Firearms and violence: A critical review, National Academies Press, 152–200. 12. Robert D. Putnam, 2000, Bowling alone: The collapse and revival of American community, Simon and Schuster. 13. CDC Wonder, average suicide rates over the period 2008–17. 14. Across the fifty US states, the correlation coefficient is .4. 15. Anne Case and Angus Deaton, 2017, “Suicide, age, and well-being: An empirical investigation,” in David A. Wise, ed., Insights in the economics of aging, National Bureau of Economic Research Conference Report, University of Chicago Press for NBER, 307–34. 16. The fractions of the birth cohorts of 1945 and 1970 who finish a four-year degree are not very different, so these results are unlikely to be attributable to changing compositions of those with and without a bachelor’s degree between the cohorts. 17.

Succeeding With AI: How to Make AI Work for Your Business by Veljko Krunic

In 2008, a widely quoted article [152] advanced the argument that if you have all the data so that you can predict what’s going to happen, does it matter why? As a corollary of that approach, theory (and theories about causality) was declared to matter much less. 3 Here, I use the word correlation to mean that there’s some form of relationship (or co-occurrence) between events. I’m not implying that AI would necessarily calculate a correlation coefficient, only that AI would learn that these events occur together. 204 CHAPTER 8 AI trends that may affect you You never have all the data Unless you possess special skills such as clairvoyance, you’ll never have all the data about all possible outcomes! To start with, you’re missing data about results that have yet to occur.

pages: 387 words: 119,409

Work Rules!: Insights From Inside Google That Will Transform How You Live and Lead by Laszlo Bock

Hunter, “The Validity and Utility of Selection Methods in Personnel Psychology: Practical and Theoretical Implications of 85 Years of Research Findings,” Psychological Bulletin 124, no. 2 (1998): 262–274. The r2 values presented in this chapter are calculated based on the reported corrected correlation coefficients (r). 86. Phyllis Rosser, The SAT Gender Gap: Identifying the Causes (Washington, DC: Center for Women Policy Studies, 1989). 87. Subsequent studies have validated the gender gap on the SAT and demonstrated racial bias as well. See, for example, Christianne Corbett, Catherine Hill, and Andresse St.

pages: 385 words: 128,358

Inside the House of Money: Top Hedge Fund Traders on Profiting in a Global Market by Steven Drobny

Risk management systems based on historical prices are one way to look at risk but are in no way faultless. Financial market history is filled with theoretically low probability or fat tail events. In LTCM’s case, its risk systems calculated roughly a 1-in-6-billion chance of a major blowup. Ironically, however, one correlation the brilliant minds of LTCM neglected to consider was the correlation coefficient of positions that were linked for no other reason than the fact 2.50 AAA Spread BAA Spread Yield (%) 2.00 Spreads Blow Out 1.50 1.00 0.50 4 Ju l-9 4 Oc t-9 4 Ja n95 Ap r-9 5 Ju l-9 5 Oc t-9 5 Ja n96 Ap r-9 6 Ju l-9 6 Oc t-9 6 Ja n97 Ap r-9 7 Ju l-9 7 Oc t-9 7 Ja n98 Ap r-9 8 Ju l-9 8 Oc t-9 8 Ja n99 Ap r-9 9 Ju l-9 9 Oc t-9 9 -9 Ap r Ja n- 94 0.00 FIGURE 2.13 Corporate Spreads to Treasuries, 1994–1999 Source: Bloomberg. 26 INSIDE THE HOUSE OF MONEY GREENSPAN ON LTCM How much dependence should be placed on financial modeling, which, for all its sophistication, can get too far ahead of human judgment?

How I Became a Quant: Insights From 25 of Wall Street's Elite by Richard R. Lindsey, Barry Schachter

Our paper examined a corollary of their result: In the presence of money illusion, the correlation between stock and bond returns will be abnormally high during periods of high inflation. For the United States, it was shown that inflation had exactly this effect on stock/bond correlations during the postwar era. As a result, asset allocation strategies that are based on the high correlation coefficients calculated using data from the 1970s and early 1980s can be expected to generate inefficient portfolios in regimes of low inflation. JWPR007-Lindsey 82 May 7, 2007 16:44 h ow i b e cam e quant Ray LeClair and I wrote a paper, “Revenue Recognition Certificates: A New Security,” in which we explored the concept and potential benefits of a new type of security.14 This security provides returns as a specified function of a firm’s sales or gross revenues over a defined period of time, say 10 years, and then expires worthless.

pages: 303 words: 67,891

Advances in Artificial General Intelligence: Concepts, Architectures and Algorithms: Proceedings of the Agi Workshop 2006 by Ben Goertzel, Pei Wang

The ANEW list contains 1,034 words, 479 of which were found in the English core. The scatter plot of our PC #1 versus the first dimension of ANEW, which is the mean value of pleasure, is represented in Figure 7. The plot shows strong correlation, with similar bimodal distributions in both PC #1 and the ANEW-pleasure dimensions. Pearson correlation coefficient r = 0.70. Figure 7. Scatter plot demonstrating strong correlation of PC #1 with the first dimension of ANEW: pleasure. The dashed line is a linear fit. The two clusters (“positive” and “negative”) are separated in each dimension. How can we match PCs with ANEW dimensions? Our correlation analysis shows that PC #1 is the best match (i.e., most highly correlated among all PCs) for ANEWpleasure, and vice versa (r = 0.70, p = 10-70).

Mastering Private Equity by Zeisberger, Claudia,Prahl, Michael,White, Bowen, Michael Prahl, Bowen White

We spent a lot of time doing educational presentations for trustees and their consultants at offsite retreats, board meetings and pension conferences. During the 1980s, our hard work finally began to pay off. As we had actual data going back to 1972, we became pension funds’ source of information on expected returns, standard deviations and correlation coefficients for the private equity “asset class.” The new term “asset class” implied a transition from a niche activity to something that was becoming institutional. We took the lead in establishing the first industry performance benchmarks, chaired the committee that established the private equity valuation guidelines, and worked with the CFA Institute to establish the guidelines for private equity performance reporting.

Beautiful Data: The Stories Behind Elegant Data Solutions by Toby Segaran, Jeff Hammerbacher

(This relationship holds even if one controls for other predictors of roll call voting, such as nominee quality and ideological distance between the senator and the nominee.) The beauty of this graph is that it combines raw data with a simple inferential model in a single plot. Typically, bivariate relationships are presented in tabular form; in this example, doing so would require either nine correlation coefficients or regression coefficients and standard errors from nine regression models, which would be ungainly, make it difficult to visualize the relationship between opinion and voting for each nominee, and create difficulties in making comparisons across nominees. The only actual numbers we include BEAUTIFUL POLITICAL DATA Download at Boykma.Com 329 Pr(Voting Yes) Bork Rehnquist 1 1 .75 .75 .75 .5 .5 .5 .25 .25 .25 42−58 0 65−33 0 40 45 50 55 60 65 45 50 55 60 65 70 Pr(Voting Yes) 55 1 .75 .75 .75 .5 .5 .25 0 78−22 0 70 75 .25 65 Ginsburg 70 75 80 85 Breyer 70 75 80 85 90 O'Connor 1 1 .75 .75 .75 .5 .5 .5 .25 .25 .25 87−9 0 70 75 80 85 State Support for Nominee 80 90−9 65 1 96−3 75 0 60 0 70 .5 .25 52−48 65 Souter 1 65 60 Roberts 1 60 58−42 0 Thomas Pr(Voting Yes) Alito 1 99−0 0 70 75 80 85 90 88 90 92 State Support for Nominee All Nominees Pr(Voting Yes) 1 .75 .5 .25 0 40 50 60 70 80 90 State Support for Nominee F I G U R E 1 9 - 4 .

pages: 386

Good Money: Birmingham Button Makers, the Royal Mint, and the Beginnings of Modern Coinage, 1775-1821 by George Anthony Selgin

Manufactured Copper Prices and Halfpenny Token Weights, 1787-1800 Year 1787 1788 1789 1790 1791 1792 1793 1794 1795 1796 1797 1798 1799 1800 Price of copper (d/lb) (Grenfell) (Tooke) 11 11 11 11 11 12 13 13 13 13 1.4 14 15 17 9.480 9.600 9.600 10.080 10.400 11.460 13.230 13.152 13.152 13.776 14.400 14.400 15.600 18.000 Average weight (ounces) Average "intrinsic value" (pence) (Tooke) (Grenfell) 0.499 0.499 0.452 0.448 0.450 0.419 0.372 0.346 0.339 0.334 0.351 0.370 0.361 0.275 0.296 0.299 0.271 0.282 0.293 0.300 0.307 0.285 0.279 0.287 0.316 0.333 0.352 0.309 0.343 0.343 0.311 0.308 0.309 0.314 0.302 0.281 0.275 0.271 0.307 0.324 0.338 0.292 12.595 0.394 Average 12.786 Correlation coefficients: -0.879 (Grenfell) -0.928 (Tooke) 0.301 0.309 SOUT(:es: Token weight~: Elks 2005. Copper prices: Thomas Tooke 1838,400 (average of reported quarterly prices); Grenfel1 1814. 146 GOOD MONEY well) have devoted so much effort and time to distinguishing specious tokens and mules from authentic issues and to documenting variants of authentic issues.

Data Wrangling With Python: Tips and Tools to Make Your Life Easier by Jacqueline Kazil

We want to determine whether perceived government corruption and child labor rates are related. The first tool we’ll use is a simple Pearson’s correlation. agate is at this point in time working on building this correlation into the agate-stats library. Until then, you can correlate using numpy. Correlation coefficients (like Pearson’s) tell us if data is related and whether one variable has any effect on another. If you haven’t already installed numpy, you can do so by running pip install numpy. Then, calculate the correlation between child labor rates and perceived government corruption by running the following line of code: import numpy numpy.corrcoef(cpi_and_cl.columns['Total (%)'].values(), cpi_and_cl.columns['CPI 2013 Score'].values())[0, 1] We first get an error which looks similar to the CastError we saw before.

The Origins of the Urban Crisis by Sugrue, Thomas J.

“Ghetto” tracts, in contrast, were poorer—only five out of twenty-six tracts with a majority-black population in 1940 had incomes above the average for all blacks. “Infill” tracts (containing second-wave black newcomers) were split evenly between above- and below-average income. To offer a more precise statistical measure of impressionistic evidence about black residential stratification, the correlation coefficient (Pearson) was calculated for tract of percentage black population in 1940 and percentage change of black population with income in 1950. The results demonstrate a clear negative correlation between the income and percentage black in 1940 and income and increase in black population. Both correlations underscore the fact that transitional tracts—those that had smaller black populations in 1940 than in 1950, and those that gained a large number of blacks between 1940 and 1950—were those tracts that had the highest median incomes. 36.

pages: 819 words: 181,185

Derivatives Markets by David Goldenberg

We are assured by replication (and no-arbitrage, of course) that, H1=C1–Δ*S1=rB and therefore, H0=C0–Δ*S0=B. 17.4.1 Volatility of the Hedge Portfolio We now want to look at the volatility of the hedged portfolio, H, in another way in terms of its components. We write it generically as H=C–Δ*S and calculate its variance using our rule from portfolio analysis that says, where and ρX,Y is the correlation coefficient between X and Y defined by ρX,Y≡Cov(X,Y)/(σX*σY). Applying this rule to our hedge portfolio H we obtain, The interpretation of is the variance of the dollar returns on the option. Similarly, is the variance of the dollar returns on the underlying stock. Why dollar returns? We will demonstrate this shortly and also formulate the analysis in terms of percentage returns to the option, the stock, and the hedge.

pages: 612 words: 187,431

The Art of UNIX Programming by Eric S. Raymond

In his paper, Graham noted accurately that computer programmers like the idea of pattern-matching filters, and sometimes have difficulty seeing past that approach, because it offers them so many opportunities to be clever. Statistical spam filters, on the other hand, work by collecting feedback about what the user judges to be spam versus nonspam. That feedback is processed into databases of statistical correlation coefficients or weights connecting words or phrases to the user's spam/nonspam classification. The most popular algorithms use minor variants of Bayes's Theorem on conditional probabilities, but other techniques (including various sorts of polynomial hashing) are also employed. In all these programs, the correlation check is a relatively trivial mathematical formula.

The Impact of Early Life Trauma on Health and Disease by Lanius, Ruth A.; Vermetten, Eric; Pain, Clare

Trauma and FM-affiliated scientists responded quite differently to the Rind study. Trauma researchers criticized the methods and conclusions of the work in multiple ways, reminding readers that estimates of psychopathology based on college samples are likely to be skewed, criticizing the use and interpretation of the correlation coefficient as the measure of effect size, objecting to biases in sampling that they believed were present, and criticizing the conclusion of “no harm” when only specific harms were assessed [30–33]. Further, they disagreed with suggestion by Rind and colleagues that the label “child sexual abuse” should be reserved for those children who were showing present symptoms and who did not “consent,” typically arguing that it is not meaningful to speak of a “willing” 5-year-old child in the context of sexual activity or to attempt “value-neutral” discussion of child abuse sexuality [31,32].

pages: 1,164 words: 309,327

Trading and Exchanges: Market Microstructure for Practitioners by Larry Harris

For comparison, the R2 of the simple market-adjusted return model is 81 percent when the portfolio standard deviation is 16 percent, and the market-adjusted return standard deviation is 7.0 percent {0.81 = (0.9)2 = (162 - 72).162}. (In the simple market-adjusted return model, the R2 is equal to the square of the correlation coefficient of the portfolio returns with the market returns.) In principle, analysts could construct stronger tests if they knew more about a manager’s presumed skill. For example, suppose an analyst believes that a manager may be skilled only in rising markets but not in falling markets. This information would allow the analyst to construct a stronger test of whether the manager is skilled.

The Art of Computer Programming: Sorting and Searching by Donald Ervin Knuth

Compare the number of inversions of h(a) — ol\X\OL2X2 • ¦ -CtkXk to inv(a); in this construction the number an does not appear in h(a).] b) Use / to define another one-to-one correspondence g having the following two properties: (i) ind(g(a)) = inv(a); (ii) inv(g(a)) — ind(a). [Hint: Consider inverse permutations.] 26. [M25] What is the statistical correlation coefficient between the number of inver- inversions and the index of a random permutation? (See Eq. 3.3.2-B4).) 27. [M37] Prove that, in addition to A5), there is a simple relationship between inv(ai a-2 ¦ ¦ ¦ an) and the n-tuple (gi, 92, • ¦ ¦,qn)- Use this fact to generalize the deriva- derivation of A7), obtaining an algebraic characterization of the bivariate generating function Hn(w,z) = J2winviai a2-an)z[nd(aia2-an), where the sum is over all n!

pages: 1,351 words: 385,579

The Better Angels of Our Nature: Why Violence Has Declined by Steven Pinker

Canadians kill at less than a third of the rate of Americans, partly because in the 19th century the Mounties got to the western frontier before the settlers and spared them from having to cultivate a violent code of honor. Despite this difference, the ups and downs of the Canadian homicide rate parallel those of their neighbor to the south (with a correlation coefficient between 1961 and 2009 of 0.85), and it sank almost as much in the 1990s: 35 percent, compared to the American decline of 42 percent.132 The parallel trajectory of Canada and the United States is one of many surprises in the great crime decline of the 1990s. The two countries differed in their economic trends and in their policies of criminal justice, yet they enjoyed similar drops in violence.