-
Notifications
You must be signed in to change notification settings - Fork 0
/
GoogleTrends.Rmd
364 lines (303 loc) · 12.6 KB
/
GoogleTrends.Rmd
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
---
title: "Using Google Trends Data to Predict Market Volatility"
author: "Oliver Jin and Santiago Tellez"
date: "2/21/2019"
output: pdf_document
---
# Introduction
## Background
## Objectives
# Previous Google Trends Analyses
https://editorialexpress.com/cgi-bin/conference/download.cgi?db_name=SNDE2018&paper_id=100
https://rstudio-pubs-static.s3.amazonaws.com/191657_7a946b1316274edfab270b0500190581.html
```{r, echo=FALSE, warning=FALSE, message=FALSE}
if (!require("pacman")) install.packages("pacman")
pacman::p_load('stringr',
'knitr',
'ggplot2',
'astsa',
'gtrendsR',
'e1071',
'randomForest')
lagpad <- function(x, k) {
if (!is.vector(x))
stop('x must be a vector')
if (!is.numeric(x))
stop('x must be numeric')
if (!is.numeric(k))
stop('k must be numeric')
if (1 != length(k))
stop('k must be a single number')
c(rep(NA, k), x)[1 : length(x)]
}
```
```{r, echo=FALSE}
#Loading Bloomberg Data
opts_knit$set(root.dir = normalizePath("../"))
SPX <- read.csv("data/SPX.csv")
VIX <- read.csv("data/VIX.csv")
VIX$Dates <- as.Date(as.character(VIX$Dates), "%m/%d/%y")
SPX$Dates <- as.Date(as.character(SPX$Dates), "%m/%d/%y")
colnames(VIX) <- c("Date", paste0(colnames(VIX)[-1], rep("_VIX", 3)))
colnames(SPX) <- c("Date", paste0(colnames(SPX)[-1], rep("_SPX", 4)))
```
```{r, echo=FALSE}
pull_data <- function(words){
start_dates <- c(seq.Date(from = as.Date("2004-01-01"), to = as.Date("2019-01-01"), by = "month"))
start_dates <- start_dates[seq(1, length(start_dates), by = 4)]
hits <- as.data.frame(matrix(NA,
nrow = as.numeric(as.Date("2019-01-01") - as.Date("2004-01-01") + 1),
ncol = length(words) + 1))
for (word in 1:length(words)) {
search_data <- gtrends(words[word], time = paste(start_dates[1], start_dates[2]))$interest_over_time
search_data$hits <- gsub("<1", "0.5", search_data$hits)
search_data$hits <- as.numeric(search_data$hits)
for (s_date in 2:(length(start_dates) - 1)) {
time_text = paste(start_dates[s_date], start_dates[s_date + 1])
temp_data <- gtrends(words[word], time = time_text)$interest_over_time
temp_data$hits <- gsub("<1", "0.5", temp_data$hits)
temp_data$hits <- as.numeric(temp_data$hits)
ratio <- temp_data$hits[1] / search_data$hits[nrow(search_data)]
search_data$hits <- search_data$hits * ratio
search_data <- rbind(search_data, temp_data[-1, ])
}
hits[,word + 1] <- search_data$hits
}
hits[,1] <- search_data$date
colnames(hits) <- c("date", words)
return(hits)
}
```
```{r, echo=FALSE}
#list of 107 words
keywords <- unique(c("stock", "market", "bond", "investment", "invest", "volatility", "gdp", "gnp",
"sp 500", "s&p 500", "spx", "vix", "russell 2000", "russell 1000", "leverage", "growth stock",
"interest rate", "interest", "debt", "asset", "liability", "etf", "accounting", "finance",
"repayment rate", "money market", "portfolio", "real estate", "treasury", "junk bond",
"recession", "revenue", "cost", "savings", "trade", "treasury bill", "risk", "risky",
"hedge fund", "investment bank", "corporate bank", "commercial bank", "bank", "bear", "bull",
"dividend", "taxes", "tax", "inflation", "recession", "risk", "crisis", "mortgage",
"credit rating", "shares", "cash", "currency", "cash inflow", "cash outflow", "bull market",
"bear market", "loan", "long term loan", "short term loan", "collateral", "gold", "silver",
"diamond", "platinum", "bronze", "gold price", "silver price","diamond price", "platinum price",
"bronze price", "water", "price", "bill", "gas", "gas price",
"gas station", "electricity", "electricity fee", "electricity bill", "food", "food price",
"grocery", "grocery price", "microsoft", "microsoft open price", "microsoft close price",
"apple", "AAPL", "amazon", "AMZN", "XOM", "closing price", "opening price", "close price", "open price",
"finance", "refinance", "mortgage loan", "home loan", "house for sale",
"apartment for sale", "house for rent", "apartment for rent", "house price", "apartment price",
"sugar", "sugar price", "salt", "salt price", "merger", "crash", "fiscal", "rates", "bubble"))
condensed_keywords <- c("stock", "crash", "bubble", "loan", "interest", "money", "bear", "bull", "risk")
results <- pull_data(condensed_keywords)
results$date <- as.Date(results$date)
```
```{r, echo=FALSE}
#Loading Words
opts_knit$set(root.dir = normalizePath("../"))
results <- read.csv("word.csv")
results$Date <- as.Date(as.character(results$Date), format = "%m/%d/%Y")
colnames(results)[1] <- "date"
results <- results[ , colSums(is.na(results)) == 0]
```
# Smoothing Words
```{r, echo=FALSE}
```
# Exploratory Data Analysis
```{r, echo=FALSE, message=FALSE, warning=FALSE}
smoother <- "3RSR"
search_results <- ggplot(data = results) +
geom_line(aes(date, smooth(stock, kind = smoother), color = "stock")) +
geom_line(aes(date, smooth(crash, kind = smoother), color = "crash")) +
geom_line(aes(date, smooth(bubble, kind = smoother), color = "bubble")) +
xlab("Date") +
ylab("Hits") +
ggtitle("Search Volume") +
theme_classic() +
theme(plot.title = element_text(hjust = 0.5), legend.position = "bottom") +
scale_colour_manual("",
breaks = c("stock", "crash", "bubble"),
values = c("blue", "green", "red"))
search_results
```
```{r, echo=FALSE, message=FALSE, warning=FALSE}
Magnitudes <- as.data.frame(abs((VIX$PX_LAST[-1] - VIX$PX_LAST[-nrow(VIX)])))
Magnitudes$date <- VIX$Date[-1]
colnames(Magnitudes) <- c("magnitude", "date")
SPX_VIX <- merge(VIX, SPX, by = "Date")
SPX_VIX_Gtrends <- merge(SPX_VIX, results, by.x = "Date", by.y = "date")
Data <- merge(Magnitudes, results)
VIX_plot <- ggplot(data = Data) +
geom_line(aes(date, smooth(log(magnitude))), color = "blue") +
xlab("Date") +
ylab("Magnitude") +
ggtitle("VIX Daily Movements") +
theme_classic() +
theme(plot.title = element_text(hjust = 0.5))
VIX_plot
```
# Models
```{r, echo=FALSE}
# VIX Up - 1 VIX Down - 0
Directions <- as.data.frame((VIX$PX_LAST[-1] > VIX$PX_LAST[-nrow(VIX)]) * 1)
Directions$date <- VIX$Date[-1]
colnames(Directions) <- c("direction", "date")
SPX_VIX <- merge(VIX, SPX, by = "Date")
SPX_VIX_Gtrends <- merge(SPX_VIX, results, by.x = "Date", by.y = "date")
Data <- merge(Directions, results)
lags <- 100
for (lag in 1:lags) {
for (word in colnames(results[-1])) {
Data_colnames <- colnames(Data)
Data <- cbind(Data, lagpad(Data[, which(colnames(Data) == word)], lag))
colnames(Data) <- c(Data_colnames, paste0(word, as.character(lag)))
}
}
Data <- na.omit(Data)
Data <- Data[, -which(colnames(Data) %in% colnames(results[-1]))]
Data1 <- Data[1:1500, ]
Data2 <- Data[1501:nrow(Data), ]
s <- svm(direction ~ ., data = Data1)
rf <- randomForest(Data1[, -c(1, 2)], as.factor(Data1[, 2]))
sum(round(predict(s, Data2)) == Data2$direction) / nrow(Data2)
sum(predict(rf, Data2) == Data2$direction) / nrow(Data2)
sum(Data2$direction == 0) / nrow(Data2)
```
```{r, echo=FALSE}
lags <- 10
for (lag in 1:lags) {
for (word in colnames(results[-1])) {
Data_colnames <- colnames(Data)
Data <- cbind(Data, lagpad(Data[, which(colnames(Data) == word)], lag))
colnames(Data) <- c(Data_colnames, paste0(word, as.character(lag)))
}
}
Data <- na.omit(Data)
Data <- Data[, -which(colnames(Data) %in% colnames(results[-1]))]
Data1 <- Data[1:2000, ]
Data2 <- Data[2001:nrow(Data), ]
training<-Data1[,-1]
```
```{r, echo=FALSE}
library(caret)
set.seed(123)
#lasso
lasso_fit <- train(magnitude ~ ., data = training,
method = "glmnet", metric = "RMSE")
lasso_RMSE<-RMSE(predict(lasso_fit, Data2), Data2$magnitude)
lasso_RMSE
#The final values used for the model were alpha = 0.55 and lambda = 0.07677905
#RMSE: 1.402875
#Non-convex penalized quantile regression
rqnc_fit <- train(magnitude ~ ., data = training,
method = "rqnc", metric = "RMSE")
rqnc_RMSE<-RMSE(predict(rqnc_fit, Data2), Data2$magnitude)
rqnc_RMSE
#The final values used for the model were lambda = 0.1 and penalty = MCP
#RMSE: 1.290154
#Least Angle Regression
lars_fit <- train(magnitude ~ ., data = training,
method = "lars", metric = "RMSE")
lars_RMSE<-RMSE(predict(lars_fit, Data2), Data2$magnitude)
lars_RMSE
#The final value used for the model was fraction = 0.05.
#RMSE: 1.409329
#Spike and Slab Regression
spikeslab_fit <- train(magnitude ~ ., data = training,
method = "spikeslab", metric = "RMSE")
spikeslab_RMSE<-RMSE(predict(spikeslab_fit, Data2), Data2$magnitude)
spikeslab_RMSE
#The final value used for the model was vars = 38.
#RMSE: 1.446507
#Boosted Generalized Additive Model
gamboost_fit <- train(magnitude ~ ., data = training,
method = "gamboost", metric = "RMSE")
gamboost_RMSE<-RMSE(predict(gamboost_fit, Data2), Data2$magnitude)
gamboost_RMSE
#The final values used for the model were mstop = 50 and prune = no.
#RMSE: 1.381269
#Stochastic Gradient Boosting
gbm_fit <- train(magnitude ~ ., data = training,
method = "gbm", metric = "RMSE")
gbm_RMSE<-RMSE(predict(gbm_fit, Data2), Data2$magnitude)
gbm_RMSE
#The final values used for the model were n.trees = 50, interaction.depth = 1, shrinkage = 0.1 and n.minobsinnode = 10.
#RMSE: 1.512827
#Elasticnet
enet_fit <- train(magnitude ~ ., data = training,
method = "enet", metric = "RMSE")
enet_RMSE<-RMSE(predict(enet_fit, Data2), Data2$magnitude)
enet_RMSE
#The final values used for the model were fraction = 0.05 and lambda = 0.1.
#RMSE: 1.430649
#Gaussian Process
gaussprLinear_fit <- train(magnitude ~ ., data = training,
method = "gaussprLinear", metric = "RMSE")
gaussprLinear_RMSE<-RMSE(predict(gaussprLinear_fit, Data2), Data2$magnitude)
gaussprLinear_RMSE
#RMSE: 2.343732
#Gaussian Process with Radial Basis Function Kernel
gaussprRadial_fit <- train(magnitude ~ ., data = training,
method = "gaussprRadial", metric = "RMSE")
gaussprRadial_RMSE<-RMSE(predict(gaussprRadial_fit, Data2), Data2$magnitude)
gaussprRadial_RMSE
#Tuning parameter 'sigma' at a value of 0.001017136.
#RMSE: 1.232594
#k-Nearest Neighbors
knn_fit <- train(magnitude ~ ., data = training,
method = "knn", metric = "RMSE")
knn_RMSE<-RMSE(predict(knn_fit, Data2), Data2$magnitude)
knn_RMSE
#The final value used for the model was k = 9.
#RMSE: 1.30769
#Model Averaged Neural Network
avNNet_fit <- train(magnitude ~ ., data = training,
method = "avNNet", metric = "RMSE")
avNNet_RMSE<-RMSE(predict(avNNet_fit, Data2), Data2$magnitude)
avNNet_RMSE
#The final values used for the model were size = 1, decay = 0.1 and bag = FALSE.
#RMSE: 1.243323
#Principal Component Analysis
pcr_fit <- train(magnitude ~ ., data = training,
method = "pcr", metric = "RMSE")
pcr_RMSE<-RMSE(predict(pcr_fit, Data2), Data2$magnitude)
pcr_RMSE
#The final value used for the model was ncomp = 2.
#RMSE: 1.273429
#CART
rpart_fit <- train(magnitude ~ ., data = training,
method = "rpart", metric = "RMSE")
rpart_RMSE<-RMSE(predict(rpart_fit, Data2), Data2$magnitude)
rpart_RMSE
#The final value used for the model was cp = 0.1130472.
#RMSE: 1.245966
#Independent Component Regression
icr_fit <- train(magnitude ~ ., data = training,
method = "icr", metric = "RMSE")
icr_RMSE<-RMSE(predict(icr_fit, Data2), Data2$magnitude)
icr_RMSE
#The final value used for the model was n.comp = 3.
#RMSE: 1.383495
#plot of the first 100 days of test set
first<-predict(gaussprRadial_fit, Data2)[1:100]
second<-predict(avNNet_fit, Data2)[1:100]
actualmagnitude<-Data2$magnitude[1:100]
a<-cbind(first,second)
colnames(a)<-c('Gaussian Process with Radial Basis Function Kernel', 'Model Averaged Neural Network')
date<-Data2$date[1:100]
ggplot() + geom_line(aes(x=date,y=actualmagnitude,colour='actual')) +
geom_line(aes(x=date,y=first,colour = 'prediction')) +
ylab('magnitude')+xlab('date') +
ggtitle("Magtitude of actual vs. best prediction") +
theme_classic() +
theme(plot.title = element_text(hjust = 0.5), legend.position = "bottom") +
scale_colour_manual("",
breaks = c("actual", "prediction"),
values = c("black", "red"))
#most significant variables
varimp_mars <- varImp(gaussprRadial_fit)
ImpMeasure<-data.frame(varImp(gaussprRadial_fit)$importance)
ImpMeasure$Vars<-row.names(ImpMeasure)
a<-ImpMeasure[order(-ImpMeasure$Overall),][1:10,]
```
# Trading Strategy
# Conclusion