Haas MFE SAS Workshop Lecture 3:

1
Haas MFE SAS WorkshopLecture 3

• Peng Liu http//faculty.haas.berkeley.edu/peliu/c
  omputing

Haas School of Business, Berkeley, MFE 2006
2
Commonly used PROCeduresin Financial Economics

• Peng Liu http//faculty.haas.berkeley.edu/peliu/c
  omputing

Haas School of Business, Berkeley, MFE 2006
3
Basic Statistical Analysis

• Univariate statistics
• PROC MEANS
• PROC UNIVARIATE
• PROC FREQ
• Bivariate and Multivariate Statistics
• PROC CORR
• PROC NPAR1WAY
• PROC TTEST

4
Comparison of PROC MEANSand PROC UNIVARIATE

• PROC MEANS
• DESCRIPTIVE STATISTICS
• CLM CSS CV KURTOSIS LCLM MAX MEAN MIN N NMISS
  RANGE SKEWNESS STD STDERR SUM SUMWGT UCLM USS VAR
• QUANTILE STATISTICS
• MEDIANP50 Q1P25 Q3P75 P1 P5 P10 P90 P95 P99
  RANGE
• HYPOTHESIS TESTING
• PROBT T

• PROC UNIVARIATE
• DESCRIPTIVE STATISTICS
• CSS CV KURTOSIS MAX MEAN MIN MODE N NMISS RANGE
  SKEWNESS STD STDMEAN SUM SUMWGT USS VAR
• QUANTILE STATISTICS
• MEDIAN P1 P5 P10 P90 P95 P99 Q1 Q3 RANGE
• QUANTILE STATISTICS
• NORMAL PROBN MSIGN PROBM SIGNRANK PROBS T PROBT
• ROBUST STATISTICS
• GINI MAD QN SN STD_SINI STD_MAD STD_QN
  STD_QRANGE STD_SN

5
PROC MEANS

• PROC MEANS DATAmfe.loan
• VAR appraisal ltv
• CLASS state
• RUN

PROC MEANS DATAmfe.loan max min VAR appraisal
ltv OUTPUT OUTm maxmaxvalue maxltv
minminvalue minltv RUN

• The default output for PROC MEANS are variable
  label N Mean Std Dev Min max
• median min max clm alpha0.05 are examples of
  options you can specify.
• You can get summary statistics for many variables
• CLASS statements will produce summary stat for
  each grouping class.
• You can suppress print using NOPRINT option
• You can save the result in a self-defined sas
  dataset.

6
PROC UNIVARIATE

• PROC UNIVARIATE DATAmfe.loan
• VAR ltv ID id
• RUN

PROC UNIVARIATE DATAmfe.loan VAR ltv
HISTOGRAM QQPLOT /normal RUN

• Use VAR to specify which variable you want to
  analyze, otherwise, this PROC will produce all
  variables
• Use ID to identify Extreme Observations, without
  ID statement it will use observation number by
  default
• Can plot histogram, quantile-quantile plots etc.
• Can do twosided T test, etc.

7
PROC FREQ

• PROC FREQ DATAmfe.loan
• TABLE term
• RUN

PROC FREQ DATAmfe.loan TABLE state
stateterm/nocol norow RUN

• One-way v.s two-way frequency table
• /CHISQ or /BINOMIAL option can be used to test
  equal proportion
• In one TABLE statement, you can produce more than
  one frequency tables
• You can suppress col percentage or/and row
  percentage by option /nocol norow

8
PROC CORR

• PROC CORR DATAmfe.loan
• VAR rate ltv fico_orig
• RUN

PROC CORR DATAmfe.loan COV SPEARMAN VAR rate
ltv fico_orig RUN

• The CORR procedure computes Pearson correlation
  coefficients, three nonparametric measures of
  association (Spearman rank-oder correlation,
  Kendalls taub and Hoeffdings measure of
  dependence D), and the probabilities associated
  with these statistics for numeric variables
• The default is Pearson correlation.
• COV option evolke the computation of covariance

9
PROC TTEST

• DATA
• INPUT a b _at__at_
• DATALINES
• 51 55 64 61 75 74 86 90
• 95 93 68 71 73 72 90 95
• RUN

PROC TTEST PAIRED ab RUN

• DATA step will produce automatic dataset, if user
  did not specify one.
• _at__at_ in INPUT lets SAS continuously read from
  datelines
• DATALINES is a SAS statement followed by lines
  of raw data.
• Data are typed continuously separated by blank,
  you can separated into a different line in the
  way you like.
• should be stand by itself
• PROC step will perform specified procedure on
  current dataset in working directory if user did
  not specify a particular dataset name
• Paired T-Test

10
PROC NPAR1WAY

• PROC NPAR1WAY DATAmfe.loan
• CLASS state
• VAR ltv
• RUN

• NONPARAMETRIC TEST FOR DIFFERENCE ACROSS ONE-WAY
  CLASSIFICATION.
• IF the normality assumption does not hold, we may
  use some nonparametric tests.
• PROC NPAR1WAY performs nonparametric tests for
  location and scale differences across a one-way
  classiication, based on the following scores
  Wilcoxin, Median, Van Der Waerden, Savage,
  Siegel-Tukey, Ansari-Bradley, Klotz, and Modd
  Scores.

11
Financial Econometrics using SAS

• Linear Models (OLS, GLS and their variates)
• PROC REG
• PROC GLM (Skip)
• Logistic Regression
• PROC LOGISTIC
• PROC GENMOD
• Hazard Regression (Cox-P.H.)
• PROC PHREG

12
Linear Model Theory

• Data (yi, xi(xi1, xi2, xik)) for i1, , n
  and yi ? R
• Model yi ?0?-1xi1 ?kxik ?i for i1,,n
• For short yX??
• where
• Assumption ?i are i.i.d. normal N(0,?2)
• Ordinary Least Square Estimation
• ? (XTX)-1XTy

13
PROC REG

• PROC REG is a SAS procedure for simple or
  multivariate linear regression models with
  continuous dependent variables.
• Part of SAS/STAT
• Model fitting (parameters, residuals, confidence
  limits, influential statistics, etc)
• Model selection (forward, backward, stepwise,
  ,etc)
• Hypothesis testing
• Model diagnostics
• Plotting
• Outputting estimates and statistics

14
PROC REG Examples

• PROC REG DATAmfe.loan
• MODEL ltv rate
• PLOT ltv rate
• QUIT

MODEL ltv rate fico_orig OLSMODEL ltv term
rate fico_orig MODEL ltv rate fico_orig
term/SELECTIONF

• Begin with PROC REG end with QUIT
• Multiple independent , dependent variables are
  separated by space
• Label OLS is optional, useful for multiple
  MODEL statement in one PROC REG
• By default, a constant is included
• Use /Options to request additional stat or
  specify model selection method
• PLOT creates a scatter plot of your regression
  data and automatically adds the regression line.

15
Logistic Regression Theory

• Data (yi, xi(xi1, xi2, xik)) for i1, , n
  and yi is a binary or ordinal response variable.
  e.g. yi ? 0,1
• Model
• Maximum Likelihood estimate of ?
• Assumption binomial Variation

16
Logistic Regression SAS procedure

• SAS has several procedures that performs logistic
  regression, e.g. GENMOD, CATMOD and LOGISTIC
• PROC LOGISTIC
• Works for binary or ordinal response variables
• Performs MLE using different optimization
  algorithms
• 4 model selection methods F, B, Stepwise, Score
• Outputs statistics to dataset
• Tests linear hypotheses of parameters

17
PROC LOGISTIC Examples

• PROC LOGISTIC DATAmfe.loan
• CLASS state edu
• MODEL default ltv age edu term rate
  state/LINKLOGIT
• RUN

• Begin with PROC LOGISTIC end with QUIT
• /LINKLOGIT option can be ignored, other options
  PROBIT, CLOGIT, CLOGLOG
• Use CLASS statement to avoid creating dummy in
  DATA step
• /option can be used to request additional stat,
  or specify selection method.
• TEST statement

18
Survival Analysis Background 1
19
Survival Analysis Background 2
20
Cox Proportional Hazard Regression
21
PROC PHREG - Example

• PROC PHREG DATAmfe.loan
• MODEL loanageprepay(0) age edu race rate ltv
  fico_orig state
• RUN

• Use WHERE option to subset sample to want to
  regress
• You can define, group variables inside PHREG
  after MODEL using IF THEN ELSE
• Handling tied data /TIESEXACT, other option
  DISCRETE
• Run PHREG for different group, use BY option,
  need to sort data.
• Use CLASS statement to create dummy variables