Introduction to Computational Finance and Financial Econometrics Descriptive Statistics

Transcription

1 You can t see this text! Introduction to Computational Finance and Financial Econometrics Descriptive Statistics Eric Zivot Summer 2015 Eric Zivot (Copyright 2015) Descriptive Statistics 1 / 28

2 Outline 1 Univariate Descriptive Statistics 2 Bivariate Descriptive Statistics 3 Time Series Descriptive Statistics Eric Zivot (Copyright 2015) Descriptive Statistics 2 / 28

3 Covariance Stationarity {..., X 1,..., X T,...} = {X t } is a covariance stationary stochastic process, and each X t is identically distributed with unknown pdf f (x). Recall, E[X t ] = µ indep of t var(x t ) = σ 2 indep of t cov(x t, X t j ) = γ j indep of t cor(x t, X t j ) = ρ j indep of t Eric Zivot (Copyright 2015) Descriptive Statistics 3 / 28

4 Descriptive Statistics Observed Sample: {X 1 = x 1,..., X T = x T } = {x t } T t=1 are observations generated by the stochastic process. Descriptive Statistics: Data summaries (statistics) to describe certain features of the data, to learn about the unknown pdf, f (x), and to capture the observed dependencies in the data. Eric Zivot (Copyright 2015) Descriptive Statistics 4 / 28

5 Time Plots Line plot of time series data with time/dates on horizontal axis. Visualization of data - uncover trends, assess stationarity and time dependence Spot unusual behavior Plotting multiple time series can reveal commonality across series Eric Zivot (Copyright 2015) Descriptive Statistics 5 / 28

6 Histograms Goal: Describe the shape of the distribution of the data {x t } T t=1 Hisogram Construction: 1 Order data from smallest to largest values 2 Divide range into N equally spaced bins [ ] 3 Count number of observations in each bin 4 Create bar chart (optionally normalize area to equal 1) Eric Zivot (Copyright 2015) Descriptive Statistics 6 / 28

7 R Functions Function hist() density() Description compute histogram compute smoothed histogram Note: The density() function computes a smoothed (kernel density) estimate of the unknown pdf at the point x using the formula: ˆf (x) = 1 Tb T ( ) x xt k b t=1 k( ) = kernel function b = bandwidth (smoothing) parameter where k( ) is a pdf symmetric about zero (typically the standard normal distribution). See Ruppert Chapter 4 for details. Eric Zivot (Copyright 2015) Descriptive Statistics 7 / 28

8 Empirical Quantiles/Percentiles Percentiles: For α [0, 1], the 100 α th percentile (empirical quantile) of a sample of data is the data value ˆq α such that α 100% of the data are less than ˆq α. Quartiles: ˆq.25 = first quartile ˆq.50 = second quartile (median) ˆq.75 = third quartile ˆq.75 ˆq.25 = interquartile range (IQR) Eric Zivot (Copyright 2015) Descriptive Statistics 8 / 28

9 R Functions Function sort() min() max() range() quantile() median() IQR() summary() Description sort elements of data vector compute minimum value of data vector compute maximum value of data vector compute min and max of a data vector compute empirical quantiles compute median compute inter-quartile range compute summary statistics Eric Zivot (Copyright 2015) Descriptive Statistics 9 / 28

10 Historical Value-at-Risk Let {R t } T t=1 denote a sample of T simple monthly returns on an investment, and let $W 0 be the initial value of an investment. For α (0, 1), the historical VaR α is: $W 0 ˆq R α ˆq R α = empirical α 100% quantile of {R t } T t=1 Note: For continuously compounded returns {r t } T t=1 use: $W 0 (exp(ˆq r α) 1) ˆq r α = empirical α 100% quantile of {r t } T t=1 Eric Zivot (Copyright 2015) Descriptive Statistics 10 / 28

11 Sample Statistics Plug-In Principle: Estimate population quantities using sample statistics. Sample Average (Mean): 1 T T x t = x = ˆµ x t=1 Sample Variance: 1 T 1 T (x t x) 2 = sx 2 = ˆσ x 2 t=1 Sample Standard Deviation: s 2 x = s x = ˆσ x Eric Zivot (Copyright 2015) Descriptive Statistics 11 / 28

12 Sample Statistics cont. Sample Skewness: 1 T 1 Sample Kurtosis: 1 T 1 T (x t x) 3 /sx 3 = ŝkew t=1 T (x t x) 4 /sx 4 = kurt t=1 Sample Excess Kurtosis: kurt 3 Eric Zivot (Copyright 2015) Descriptive Statistics 12 / 28

13 R Functions Function Package Description mean() base compute sample mean colmeans() base compute column means of matrix var() stats compute sample variance sd() stats compute sample standard deviation skewness() PerformanceAnalytics compute sample skewness kurtosis() PerformanceAnalytics compute sample excess kurtosis Note: Use the R function apply(), to apply functions over rows or columns of a matrix or data.frame. Eric Zivot (Copyright 2015) Descriptive Statistics 13 / 28

14 Empirical Cumulative Distribution Function Recall, the CDF of a random variable X is: F X (x) = Pr(X x) The empirical CDF of a random sample is: ˆF X (x) = 1 n (#x i x) = number of x i values x sample size Eric Zivot (Copyright 2015) Descriptive Statistics 14 / 28

15 Empirical Cumulative Distribution Function cont. How to compute and plot ˆF X (x) for a sample {x 1,..., x n }. Sort data from smallest to largest values: {x (1),..., x (n) } and compute ˆF X (x) at these points Plot ˆF X (x) against sorted data {x (1),..., x (n) } Use the R function ecdf() Note: x (1),..., x (n) are called the order statistics. In particular, x (1) = min(x 1,..., x n ) and x (n) = max(x 1,..., x n ). Eric Zivot (Copyright 2015) Descriptive Statistics 15 / 28

16 Quantile-Quantile (QQ) Plots A QQ plot is useful for comparing your data with the quantiles of a distribution (usually the normal distribution) that you think is appropriate for your data. You interpret the QQ plot in the following way: If the points fall close to a straight line, your conjectured distribution is appropriate If the points do not fall close to a straight line, your conjectured distribution is not appropriate and you should consider a different distribution Eric Zivot (Copyright 2015) Descriptive Statistics 16 / 28

17 R Functions Function Package Description qqnorm() stats QQ-plot against normal distribution qqline() stats draw straight line on QQ-plot qqplot() car QQ-plot against specified distribution Eric Zivot (Copyright 2015) Descriptive Statistics 17 / 28

18 Outliers Extremely large or small values are called outliers Outliers can greatly influence the values of common descriptive statistics. In particular, the sample mean, variance, standard deviation, skewness and kurtosis Percentile measures are more robust to outliers: outliers do not greatly influence these measures (e.g. median instead of mean; IQR instead of SD) Eric Zivot (Copyright 2015) Descriptive Statistics 18 / 28

19 Outliers cont. IQR (interquartile range) - outlier robust measure of spread: IQR = q.75 q.25 Moderate Outlier: ˆq IQR < x < ˆq IQR ˆq.25 3 IQR < x < ˆq IQR Extreme Outlier: x > ˆq IQR x < ˆq.25 3 IQR Eric Zivot (Copyright 2015) Descriptive Statistics 19 / 28

20 Boxplots A box plot displays the locations of the basic features of the distribution of one-dimensional data the median, the upper and lower quartiles, outer fences that indicate the extent of your data beyond the quartiles, and outliers, if any. R functions Function Package Description boxplot() graphics box plots for multiple series chart.boxplot() PerformanceAnalytics box plots for asset returns Eric Zivot (Copyright 2015) Descriptive Statistics 20 / 28

21 Outline 1 Univariate Descriptive Statistics 2 Bivariate Descriptive Statistics 3 Time Series Descriptive Statistics Eric Zivot (Copyright 2015) Descriptive Statistics 21 / 28

22 Bivariate Descriptive Statistics {..., (X 1, Y 1 ), (X 2, Y 2 ),... (X T, Y T ),...} = {(X t, Y t )} Covariance stationary bivariate stochastic process with realized values: {(x 1, y 1 ), (x 2, y 2 ),... (x T, y T )} = {(x t, y t )} T t=1 Scatterplot: XY plot of bivariate data R functions: plot(), pairs() Eric Zivot (Copyright 2015) Descriptive Statistics 22 / 28

23 Bivariate Descriptive Statistics cont. Sample Covariance: 1 T 1 Sample Correlation: T (x t x)(y t ȳ) = s xy = ˆσ xy t=1 s xy s x s y = r xy = ˆρ xy Eric Zivot (Copyright 2015) Descriptive Statistics 23 / 28

24 R Functions Function Package Description var() stats compute sample variance-covariance matrix cov() stats compute sample variance-covariance matrix cor() stats compute sample correlation matrix Eric Zivot (Copyright 2015) Descriptive Statistics 24 / 28

25 Outline 1 Univariate Descriptive Statistics 2 Bivariate Descriptive Statistics 3 Time Series Descriptive Statistics Eric Zivot (Copyright 2015) Descriptive Statistics 25 / 28

26 Time Series Descriptive Statistics Sample Autocovariance: ˆγ j = 1 T 1 T t=j+1 Sample Autocorrelation: ˆρ j = ˆγ j, j = 1, 2,... ˆσ 2 (x t x)(x t j x) j = 1, 2,... Sample Autocorrelation Function (SACF): Plot ˆρ j against j Eric Zivot (Copyright 2015) Descriptive Statistics 26 / 28

27 R Functions Function Package Description acf() stats compute and plot sample autocorrelations chart.acf() PerformanceAnalytics plot sample autocorrelations chart.acfplus() PerformanceAnalytics plot sample autocorrelations Eric Zivot (Copyright 2015) Descriptive Statistics 27 / 28

28 You can t see this text! faculty.washington.edu/ezivot/ Eric Zivot (Copyright 2015) Descriptive Statistics 28 / 28