Analysis of Financial Time Series, Third Edition

Appendix B: R and S-Plus Commands for Neural Network

The following commands are used in R or S-Plus to build the 3–2–1 skip-layer feed-forward network of Example 4.7. A line starting with # denotes a comment. The data file is m-ibmln.txt. The library used is nnet.

# load the data into R or S-Plus workspace.

x_scan(file=‘m-ibmln.txt’)

# select the output: r(t)

y_x[4:864]

# obtain the input variables: r(t-1), r(t-2), and r(t-3)

ibm.x_cbind(x[3:863]_,x[2:862],x[1:861])

# build a 3-2-1 network with skip layer connections

# and linear output.

ibm.nn_nnet(ibm.x,y,size=2,linout=T,skip=T,maxit=10000,

decay=1e-2,reltol=1e-7,abstol=1e-7,range=1.0)

# print the summary results of the network

summary(ibm.nn)

# compute \& print the residual sum of squares.

sse_sum((y-predict(ibm.nn,ibm.x))ˆ2)

print(sse)

#eigen(nnet.Hess(ibm.nn,ibm.x,y),T)$values

# setup the input variables in the forecasting subsample

ibm.p_cbind(x[864:887],x[863:886],x[862:885])

# compute the forecasts

yh_predict(ibm.nn,ibm.p)

# The observed returns in the forecasting subsample

yo_x[865:888]

# compute \& print the sum of squares of forecast errors

ssfe_sum((yo-yh)ˆ2)

print(ssfe)

# quit S-Plus or R

q()

Exercises

4.1 Consider the daily simple returns of Johnson & Johnson stock from January 1998 to December 2008. The data are in the file d-jnj9808.txt or can be obtained from CRSP. Convert the returns into log returns in percentage. (a) Build a GJR model for the log return series. Write down the fitted model. Is the leverage ...