*** This file contains some Splus commands useful for Financial 
*** Time Series Analysis.

***
**** Commands associated with FinMetrics module
     [That is, commands for Splus 6.0 or higher for Windows]

Part A)   For Chapter 2 of the textbook.

Click Start > Programs > Splus 6.0 (or 6.1 to start the Splus program)
Click File > Load Module to select and download the finmetrics module
Click File > Import Data > From Files (then 
  (a) select Browse in the pop-up  window to find the data file 
  (b) specify Data Set name (of your choice, I shall use DAA as an example))
   then click OK to complete the data input.
A data window will appear with the data in columns.
The columns are named as Col1, Col2, etc.
If you like to change the name of a particular column, move the Cursor to 
that column, click with the "Right Button" of the mouse to produce a 
pop-up window, select "Properties" to produce a pop-up window. You can change 
the column name in that small window, then click OK.

(If "Command Window" already appears, skip.)
Click Window > Commands window to produce a command window

You can now issue commands in the Commands Window

For simplicity, you can define a variable for the time series 
you like to analyze. Fr instance, suppose that we like to 
analyze Col1 of the data set "DAA" and name the series "rtn". 

rtn = DAA$Col1

For descriptive statistics of the time series "rtn". 

summaryStats(rtn)  

To compute the ACF of "rtn":

acf(rtn, lag.max = 20)  [The lag.max is optional, an ACF plot will appear.]

To compute Ljung-Box Statistics of the first 10 lags

autocorTest(rtn, lag.n = 10)

To compute the first 12 lags of PACF of "rtn"

acf(rtn, lag.max=12, type='partial')

To quit

q() [Or Click File > Exit]

To obtain an Ordinary Least Squares estimation of the linear 
regression y = b0 + b1*x1 + b2*x2 + b3*b3 + error:

fit = OLS(y ~ 1 + x1 + x2 + x3)

summary(fit)   [To obtain output]

Then, use names(fit) to see the variables from the object ``fit''.

To fit the above regression model WITHOUT the constant term:

fit= OLS(y ~ -1 + x1 + x2 + x3)

sumamry(fit)

To fit an AR(3) model using OLS command for a time series ``rtn'':

fit = OLS(rtn ~ ar(3))

summary(fit)

To fit an AR(3) model using maximum likelihood method to ``rtn'': 

m.fit = arima.mle(rt.rm, model=list(order=c(3,0,0)),xreg=1)
[Note: the command "xreg=1" is used to fit a constant term in the model. 
Otherwise, the command assume the mean of the time series is zero.]

m.fit

To fit an airline model to a monthly time series ``rtn'':
(a) first, specify the model

air.md = list(list(order=c(0,1,1)),list(order=c(0,1,1),period=12))
  
(b) next, perform estimation

m.fit = arima.mle(rtn, model=air.md)

m.fit


To perform model checking using various plots.:

chk = arima.diag(m.fit)


To use the fitted model m.fit to produce 12 forecasts:

m.fore = arima.forecast(rtn, n = 12, model=m.fit$model)

m.fore

To fit the following regression model with time series error:
y(t) = b0 + b1*x(t) + b2*y(t-1) + b3*x(t-1) + error

fit = OLS(y~x + tslag(y,1) + tslag(x,1), na.rm=T)
[The command "na.rm=T" is to remove any missing values due to 
the time-series lag operator "tslag".]

summary(fit)


===================================================================
Part B). For volatility models: Chapter 3 of the textbook

Again, I use "rtn" as the name of the time series to be analyzed.

(1) To test for serial correlations in "rtn"

autocorTest(rtn, lag.n=10)

(2) To test for ARCH effects using 10 lags:

archTest(rtn, lag.n=10)

(3) To obtain PACF of squared return series:

acf(rtn^2, lag.max=12, type='partial')

(4) To fit an ARCH(2) model to "rtn":

arch2 = garch(rtn~1, ~garch(2,0))

summary(arch2)

(5) To see various plots concerning the fit:

plot(arch2)

(6) To fit a GARCH(1,1) model to "rtn":

fit = garch(rtn~1, ~garch(1,1))

sumamry(fit)

(7) To fit a GARCH(1,1) model with Student-t innovations:

fit=garch(rtn~1, ~garch(1,1), cond.dist="t")

sumamry(fit)

(8) To fit a GARCH(1,1) model with generalized error distributions:

fit = garch(rtn~1m ~garch(1,1), cond.dist="ged")

summary(fit)

(9) To fit an EGARCH(1,1) model with leverage effect and generalized 
error distribution:

fit=garch(rtn~1, ~egarch(1,1), leverage=T, cond.dist="ged")

summary(fit)

(10) To fit a GARCH(1,1)-M model using volatility:

fit=garch(rtn~var.in.mean, ~garch(1,1))

summary(fit)

(11) To fit a GARCH(1,1)-M model using standard deviation:

fit=garch(rtn~sd.in.mean, ~garch(1,1))

summary(fit)

(12) To produce 1-step to 10-step ahead forecasts:

fore=predict(fit,10)

fore

(13) To fit a threshold GARCH(1,1) model:

fit=garch(rtn~1, ~tgarch(1,1), trace=F)

summary(fit)