forked from RobertMcPherson/BayesianVectorAutoRegression
-
Notifications
You must be signed in to change notification settings - Fork 0
/
DJI_TimeSeries_BigVAR
451 lines (357 loc) · 14 KB
/
DJI_TimeSeries_BigVAR
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
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
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
## ----global_options, include=FALSE---------------------------------------
#knitr::opts_chunk$set(include=TRUE, warning=FALSE, message=FALSE,echo=FALSE)
## ----install_libraries, results='hide'-----------------------------------
#install.packages("sqldf")
#install.packages("dummies")
#install.packages("forecast")
#install.packages("orderedLasso")
#install.packages("glmnet")
#install.packages("h2o")
#install.packages("addendum")
#install.packages("testthat")
#install.packages("imputateTS")
#install.packages("BigVAR")
#install.packages("reshape")
#install.packages("ggplot2")
#install.packages("Quandl")
#install.packages("knitr")
#devtools::use_testthat
#library(devtools)
#install.packages("devtools")
#install_github("SteffenMoritz/imputeTS")
#install_github("gabrielrvsc/HDeconometrics")
#install.packages("HDeconometrics")
#install.packages("githubinstall")
#install.packages("quantmod")
rm(list=ls())
#setwd("R:/AnalyticsTeam/Personal/May/BitBucket/Gemini Time Series")
#setwd("C:/Users/rmcpherson/Documents/Segments/Phil Welt Segment/Gemini")
library(sqldf) #for running sql on data frames
library(dummies) #for creating one-hot encoding
library(forecast) #for the Holt-Winters forecast filter
library(glmnet) #for running regularized GLM
library(knitr) #for reproducible research, i.e., Markdown
#library(testthat)
library(BigVAR)
library(orderedLasso)
library(reshape)
library(ggplot2)
library(Quandl)
library(HDeconometrics)
library(imputeTS)
library(quantmod)
library(xts)
## ----set_globals---------------------------------------------------------
##########################
##Input Global Variables##
##########################
##########################
#Input the column name of the dependent variable to predict.
dependent.variable <- "DJI.Chg"
##########################
##########################
#Set the maximum lag for adjusting the variables in the data.
#each variable will get a new column for each lag, up to the maximum set here.
maxlag <- 30
##########################
##########################
#Type 'TRUE' if you want to include an offset in the GLM calculation, FALSE otherwise.
include.offset <- FALSE
##########################
##########################
#Type the column name of the variable you would like to use as an offset, if any.
#offset.variable <- "UnitCount"
##########################
##########################
#Input the column name that has the time increments in it, such as years, or year/months.
time.increment.variable <- "Date"
##########################
##########################
#Select whether to include plots with the arima, pre-whitening step
include.arima.plots <- FALSE
##########################
##########################
#Select whether to include cross correlation plots
include.cross.correlation.plots <- TRUE
##########################
##########################
#Select whether to include quartile to quartile (QQ) plots
include.QQ.plots <- FALSE
##########################
## ----load_data, results='hide'-------------------------------------------
#Note: this process takes the data in descending order, with the most recent data at the
CPI <- Quandl("RATEINF/CPI_USA", api_key="DJGcfzQc5RYP1JSycMBv", collapse="annual", start_date="1960-12-31", end_date="2017-12-31", type="raw", order="asc", force_irregular=TRUE)
#Nonfinancial corporate business; short-term debt as a percentage of total debt, Annual
shortTermDebtToLongTerm <- Quandl("FED/FL104140006_A", api_key="DJGcfzQc5RYP1JSycMBv", collapse="annual", start_date="1960-12-31", end_date="2017-12-31", type="raw", order="asc", force_irregular=TRUE)
#Financial soundness indicator, households; debt as a percent of gross domestic product
debtToGDP <- Quandl("FED/FL010000336_Q", api_key="DJGcfzQc5RYP1JSycMBv",collapse="annual", start_date="1960-12-31", end_date="2017-12-31", type="raw", order="asc", force_irregular=TRUE)
GDP <- Quandl("FED/FU086902001_A", api_key="DJGcfzQc5RYP1JSycMBv", collapse="annual", start_date="1960-12-31", end_date="2017-12-31", type="raw", order="asc", force_irregular=TRUE)
m1Velocity <- Quandl("FRED/M1V", api_key="DJGcfzQc5RYP1JSycMBv",collapse="annual", start_date="1960-12-31", end_date="2017-12-31", type="raw", order="asc", force_irregular=TRUE)
m2Velocity <- Quandl("FRED/M2V", api_key="DJGcfzQc5RYP1JSycMBv", collapse="annual", start_date="1960-12-31", end_date="2017-12-31", type="raw", order="asc", force_irregular=TRUE)
raw.ts <- cbind(m1Velocity, m2Velocity, GDP[-1,], debtToGDP[-1,], shortTermDebtToLongTerm[-1,], CPI[-1,])
raw.ts <- cbind(m1Velocity, m2Velocity, GDP[-1,], debtToGDP[-1,], shortTermDebtToLongTerm[-1,], CPI[-1,])
#save time increment vector
#time.increments <- unique(raw_data_dummies[,time.increment.variable])
#time.increments <- time.increments[sort.list(time.increments, decreasing=FALSE)]
time.increments <- raw.ts[,time.increment.variable]
rownames(raw.ts) <- raw.ts[,1]
data <- raw.ts[, !(names(raw.ts) %in% "Date")]
colnames(data) <- c("CPI","shortTermDebtToLongTerm","debtToGDP","GDP","m1Velocity","m2Velocity")
##Import birth data
births <- read.csv("BirthsModified.csv")
#impute the na values
births.interp <- na.interpolation(ts(births))
births.export <- apply(births.interp, 2, rev)
head(births.export)
write.csv(births.export, "births_interp_no_blanks.csv")
births.trunc <- births.interp[1:57,3:5]
births.2009 <- apply(births.trunc, 2, rev)
head(births.2009)
#na.interpolation(ts(births.trunc[,2]))
#births.interp <- apply(births.trunc, 2, na.interpolation)
#colnames(births.interp) <- c("Year", "Births", "BirthRate")
#head(births.interp)
#head(data)
#str(data)
#?rev
#births.2009 <- apply(births.interp[], 2, rev)
#################################
##Not included - series too short
##Get SPX data
SPX <- getSymbols("^GSPC",auto.assign = FALSE)
?getSymbols
SPX.xts <- as.xts(SPX)
SPX.yearly <- to.yearly(SPX.xts)
#getSymbols("DJI", src = "yahoo", from = start_date, to = end_date)
DJI <- getSymbols("DJI", src = "yahoo")
DJI.xts <- as.xts(DJI)
DJI.interp <- na.interpolation(DJI.xts)
DJI.yearly <- to.yearly(DJI.interp)
#################################
##Import historical Dow Jones Industrial Average data
Dow <- read.csv("DJI_Historical_Chg_Data.csv")
head(Dow)
Dow.interp <- na.interpolation(Dow)
DJI.Chg <- Dow[1:57,2]
SeriesData <- cbind(births.2009, data, DJI.Chg)
SeriesData <- SeriesData[,-1] #remove the "Year" column
head(SeriesData)
col.names <- colnames(SeriesData)
x <- SeriesData[, !(names(SeriesData) %in% dependent.variable)]
head(x)
#scale the dependent variable
x.scaled <- scale(x)
#Isolate dependent variable values, based on name given in global variable inputs above
y <- SeriesData[,dependent.variable]
y.unscaled <- y
#scale the dependent variable
y.scaled <- scale(y)
#save column names
x.colnames <- colnames(x)
## ----ARIMA, results='asis'-----------------------------------------------
#i=20
num.cols <- length(x[1,])
#apply(x,1,function(x) sum(is.na(x)))
#str(x)
#?auto.arima
#generate ARIMA plots...intent is to get ARIMA parameters, rather than forecasts
x.arima.residuals = NULL
for (i in 1:num.cols){
fit <- auto.arima(x.scaled[,i])
pdf(file=paste("plots/arima_",x.colnames[i],".pdf",sep=""))
if(include.arima.plots == TRUE){
par(mar=c(8,4,2,2))
plot(forecast(fit,h=maxlag), sub=paste(x.colnames[i]))
}
dev.off()
#assemble a table of ARIMA residuals for use in cross-correlation analysis
temp.resid <- resid(fit)
x.arima.residuals <- as.matrix(cbind(x.arima.residuals, temp.resid))
}
#run arima transformation on the dependent variable
fit=NULL
fit <- auto.arima(y.scaled)
par(mar=c(8,4,2,2))
pdf(file=paste("plots/arima_",dependent.variable,".pdf",sep=""))
plot(forecast(fit,h=1), sub=paste(dependent.variable, sep=""))
dev.off()
y.arima.residuals <- resid(fit)
## ------------------------------------------------------------------------
if(include.QQ.plots == TRUE){
#check distributions of independent variables for normality
for (i in 1:length(x.scaled[1,])){
pdf(file=paste("plots/qqnorm_",x.colnames[i],".pdf",sep=""))
qqnorm(x.arima.residuals[,i], main=paste(x.colnames[i]))
dev.off()
}
#check dependent variable for normality
pdf(file=paste("plots/qqnorm_",dependent.variable,".pdf",sep=""))
qqnorm(y.arima.residuals, main=paste(dependent.variable,sep=""))
dev.off()
}
## ------------------------------------------------------------------------
#i=2
##cross correlation analysis
#leading indicators in 'x' will have negative lag values for the most significant
#correlations in the chart.
#note: analysis is run on ARIMA residuals so as to pre-whiten the data
#i=1
dir.create("plots")
if(include.cross.correlation.plots == TRUE){
for (i in 1:length(x[1,])){
pdf(file=paste("plots/ccf_",x.colnames[i],".pdf",sep=""))
par(mar=c(5,7,4,2)) #set the margins so title does not get cut off
ccf(x.arima.residuals[,i], y.arima.residuals, plot=TRUE, main=paste(x.colnames[i]), na.action = na.contiguous)
dev.off()
}
}
## ----analytical_dataset, results='hide'----------------------------------
#x
#y
#x.arima.residuals <- apply(x.arima.residuals, 2, rev)
#y.arima.residuals <- rev(y.arima.residuals)
#time.increments <- rev(time.increments)
#x2 <- subset(x, select= -c(X.Dividends, X.From.the.rest.of.the.world.6.))
Y <- cbind.data.frame(x, y)
colnames(Y) <- col.names
nms <- dependent.variable
Y <- as.matrix(Y)
# Fit a Basic VAR-L(3,4) on simulated data
T1=floor(nrow(Y)/3)
T2=floor(2*nrow(Y)/3)
#?constructModel
#m1=constructModel(Y,p=4,struct="Basic",gran=c(20,10),verbose=FALSE,IC=FALSE,T1=T1,T2=T2,ONESE=TRUE)
#m1=constructModel(Y,p=4,struct="Tapered",gran=c(50,10),verbose=FALSE,T1=T1,T2=T2,IC=FALSE)
#plot(m1)
#results=cv.BigVAR(m1)
#plot(results)
#predict(results,n.ahead=1)
#SparsityPlot.BigVAR.results(results)
#str(results)
#results@preds
#results@alpha
#results@Granularity
#results@Structure
#results@lagmax
#results@Data
#plot(results@Data)
#install.packages("devtools")
#library(devtools)
#install_github("gabrielrvsc/HDeconometrics")
###################################
#The above, BigVAR package will not handle data sets this wide. Trying the
#Bayesian Vector Auto Regression (BVAR) algorithm
###################################
##Perform analysis on pre-whitened data
# = load package and data = #
#install.packages("HDeconometrics")
#library(HDeconometrics)
#data("voldata")
# = Break data into in and out of sample to test model accuracy= #
#Yin=voldata[1:5499,]
#Yout=voldata[-c(1:5499),]
Yin = Y[1:T2,]
Yout = Y[(T2+1):(T1+T2),]
# = Run models = #
# = OLS = #
#modelols=HDvar(Yin,p=2) # takes a while to run
#predols=predict(modelols,h=2)
# = BVAR = #
#?lbvar
#?predict
modelbvar=lbvar(Yin, p = 5, delta = 0.5)
predbvar=predict(modelbvar,h=5)
# = Forecasts of the volatility = #
k=paste(dependent.variable)
pdf(file=paste("plots/forecast.pdf",sep=""))
plot(c(Y[,k],predbvar[,k]),type="l", main=paste(dependent.variable))
#lines(c(rep(NA,length(Y[,k])),predols[,k]))
lines(c(rep(NA,length(Y[,k])),predbvar[,k]))
abline(v=length(Y[,k]),lty=2,col=4)
#legend("topleft",legend="BVAR",col=2,lty=1,lwd=1,seg.len=1,cex=1,bty="n")
dev.off()
# = Overall percentual error = #
#MAPEols=abs((Yout-predols)/Yout)*100
#MAPEbvar=abs((Yout-predbvar)/Yout)*100
#matplot(MAPEols,type="l",ylim=c(0,80),main="Overall % error",col="lightsalmon",ylab="Error %")
#aux=apply(MAPEbvar,2,lines,col="lightskyblue1")
#lines(rowMeans(MAPEols),lwd=3,col=2,type="b")
#lines(rowMeans(MAPEbvar),lwd=3,col=4,type="b")
#legend("topleft",legend=c("OLS","BVAR"),col=c(2,4),lty=1,lwd=1,seg.len=1,cex=1,bty="n")
# = Influences = #
#aux=modelbvar$coef.by.block[2:23]
#impacts=abs(Reduce("+", aux ))
#diag(impacts)=0
#I=colSums(impacts)
#R=rowSums(impacts)
#par(mfrow=c(2,1))
#barplot(I,col=rainbow(30),cex.names = 0.3, main = "Most Influent")
#barplot(R,col=rainbow(30),cex.names = 0.3, main = "Most Influenced")
pdf(file=paste("plots/barchart.pdf",sep=""))
aux=modelbvar$coef.by.block
impacts=abs(Reduce("+", aux ))
diag(impacts)=0
I=colSums(impacts)
R=rowSums(impacts)
par(mfrow=c(2,1))
barplot(I,col=rainbow(30),cex.names = 0.3, main = "Most Influent")
barplot(R,col=rainbow(30),cex.names = 0.3, main = "Most Influenced")
dev.off()
###################################
##Perform analysis on NON pre-whitened data
Y <- cbind.data.frame(x.scaled, y.unscaled)
colnames(Y) <- col.names
head(Y)
nms <- dependent.variable
Y <- as.matrix(Y)
# Fit a Basic VAR-L(3,4) on simulated data
T1=floor(nrow(Y)/3)
T2=floor(2*nrow(Y)/3)
# = Break data into in and out of sample to test model accuracy= #
Yin = Y[1:T2,]
Yout = Y[(T2+1):(T1+T2),]
# = Run models = #
# = OLS = #
#modelols=HDvar(Yin,p=2) # takes a while to run
#predols=predict(modelols,h=2)
# = BVAR = #
#?lbvar
#?predict
modelbvar=lbvar(Yin, p = 5, delta = 0.5)
predbvar=predict(modelbvar,h=5)
# = Forecasts of the volatility = #
k=paste(dependent.variable)
pdf(file=paste("plots/forecast_not_whitened.pdf",sep=""))
plot(c(Y[,k],predbvar[,k]),type="l", main=paste(dependent.variable, "Not Whitened"))
#lines(c(rep(NA,length(Y[,k])),predols[,k]))
lines(c(rep(NA,length(Y[,k])),predbvar[,k]))
abline(v=length(Y[,k]),lty=2,col=4)
#legend("topleft",legend="BVAR",col=2,lty=1,lwd=1,seg.len=1,cex=1,bty="n")
dev.off()
# = Overall percentual error = #
#MAPEols=abs((Yout-predols)/Yout)*100
#MAPEbvar=abs((Yout-predbvar)/Yout)*100
#matplot(MAPEols,type="l",ylim=c(0,80),main="Overall % error",col="lightsalmon",ylab="Error %")
#aux=apply(MAPEbvar,2,lines,col="lightskyblue1")
#lines(rowMeans(MAPEols),lwd=3,col=2,type="b")
#lines(rowMeans(MAPEbvar),lwd=3,col=4,type="b")
#legend("topleft",legend=c("OLS","BVAR"),col=c(2,4),lty=1,lwd=1,seg.len=1,cex=1,bty="n")
# = Influences = #
#aux=modelbvar$coef.by.block[2:23]
#impacts=abs(Reduce("+", aux ))
#diag(impacts)=0
#I=colSums(impacts)
#R=rowSums(impacts)
#par(mfrow=c(2,1))
#barplot(I,col=rainbow(30),cex.names = 0.3, main = "Most Influent")
#barplot(R,col=rainbow(30),cex.names = 0.3, main = "Most Influenced")
pdf(file=paste("plots/barchart_not_whitened.pdf",sep=""))
aux=modelbvar$coef.by.block
impacts=abs(Reduce("+", aux ))
diag(impacts)=0
I=colSums(impacts)
R=rowSums(impacts)
par(mfrow=c(2,1))
barplot(I,col=rainbow(30),cex.names = 0.3, main = "Most Influent")
barplot(R,col=rainbow(30),cex.names = 0.3, main = "Most Influenced")
dev.off()