-
Notifications
You must be signed in to change notification settings - Fork 0
/
model_build.Rmd
615 lines (516 loc) · 22.3 KB
/
model_build.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
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
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
---
title: "predictR"
author: "Wesley Chioh"
date: "April 19, 2020"
output: html_document
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE)
# setting up shop
library(tidyverse)
library(ggplot2)
library(eia)
library(fredr)
library(zoo)
library(plotly)
library(caret)
library(gbm)
library(doParallel)
library(stargazer)
library(rjson)
library(lubridate)
library(jsonlite)
library(ggraph)
library(igraph)
library(kableExtra)
# load API key
eia_api_key <- readtext::readtext("eia_api_key.txt") %>%
.$text
fred_api_key <- readtext::readtext("fred_api_key.txt") %>%
.$text
eia_set_key(eia_api_key)
fredr_set_key(fred_api_key)
# load datasets
gdp <- readxl::read_excel("./data/Monthly GDP for Project.xlsx")
Pull <- readxl::read_excel("./data/Pull.xlsx")
# setting seed to ensure reproducibility
set.seed(0506)
```
## Regression Analysis
```{r}
# Real GDP (Level) > Y
#Probability of Losing a Job >X3
#Job Openings per Unemployed Person >X4
#Ratio of Hires to Openings > X5
#Business Survey > X6
#Business Survey (Peak Diff) > X7
#Yield Curve > X8
#Excnomic Policy Index > X9
#Equity Uncertainty > X10
#Unemployment Rate (YoY) >X11
LOGY <- log(Pull$Y)
LOGY1 <- log(lag(Pull$Y,1))
AnnGDP <- (LOGY-LOGY1)*12
GDP_LOG <- log(Pull$Y)
# X1 Functional Form Test
X1Squared <- Pull$X1*Pull$X1
X1Root <- sqrt(Pull$X1)
TestX1 <- lm(AnnGDP~Pull$X1+X1Squared)
TestX1
stargazer(TestX1, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A1 <- (Pull$X1+X1Squared)
summary(TestX1)
# X2 Functional Form Test
X2Squared <- Pull$X2*Pull$X2
X2Root <- sqrt(Pull$X2)
TestX2 <- lm(AnnGDP~Pull$X2)
TestX2
stargazer(TestX2, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A2 <- Pull$X2
# X3 Functional Form Test
X3Squared <- Pull$X3*Pull$X3
X3Root <- sqrt(Pull$X3)
TestX3 <- lm(AnnGDP~Pull$X3)
TestX3
stargazer(TestX3, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A3 <- Pull$X3
# X4 Functional Form Test
X4Squared <- Pull$X4*Pull$X4
X4Root <- sqrt(Pull$X4)
TestX4 <-lm(AnnGDP~Pull$X4+X4Squared)
TestX4
stargazer(TestX4, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A4=(Pull$X4+X4Squared)
# X5 Functional Form Test
X5Squared <- Pull$X5*Pull$X5
X5Root <- sqrt(Pull$X5)
TestX5 <- lm(AnnGDP~Pull$X5+X5Squared)
TestX5
stargazer(TestX5, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A5 <- Pull$X5+X5Squared
# X6 Functional Form Test
X6Squared <- Pull$X6*Pull$X6
X6Root <- sqrt(Pull$X6)
TestX6 <- lm(AnnGDP~Pull$X6+X6Squared)
TestX6
stargazer(TestX6, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A6 <- Pull$X6+X6Squared
# X7 Functional Form Test
X7Squared <- Pull$X7*Pull$X7
X7Root <- sqrt(Pull$X7)
TestX7 <- lm(AnnGDP~Pull$X7+X7Squared)
TestX7
stargazer(TestX7, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A7 <- Pull$X7+X7Squared
# X8 Functional Form Test
X8Squared <- Pull$X8*Pull$X8
X8Root <- sqrt(Pull$X8)
TestX8 <- lm(AnnGDP~Pull$X8)
TestX8
stargazer(TestX8, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A8 <- Pull$X8
# X9 Functional Form Test
X9Squared <- Pull$X9*Pull$X9
X9Root <- sqrt(Pull$X9)
TestX9 <- lm(AnnGDP~Pull$X9+X9Squared)
TestX9
stargazer(TestX9, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A9 <- Pull$X9+X9Squared
# X10 Functional Form Test
X10Squared<- Pull$X10*Pull$X10
X10Root <- sqrt(Pull$X10)
TestX10 <- lm(AnnGDP~Pull$X10)
TestX10
stargazer(TestX10, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A10 <- Pull$X10
# X11 Functional Form Test
X11Squared <- Pull$X11*Pull$X11
X11Root <- sqrt(Pull$X11)
TestX11 <- lm(AnnGDP~Pull$X11)
TestX11
stargazer(TestX11, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
A11 <- Pull$X11
#First Test of All Best Form Variables
CurrentGDP <- lm(GDP_LOG~ A3+A4+A5+A6+A7+A8+A9+A10+A11)
CurrentGDP
stargazer(CurrentGDP, type="text", title="Baseline Results", single.row=TRUE,
ci=TRUE, ci.level=0.95)
# Gathering all predictors
predictors <- cbind(A3,A4,A5,A6,A7,A8,A9,A10,A11,GDP_LOG) %>%
as.data.frame()
saveRDS(predictors, "./models/predictors.RDS")
# function to automate lag calculation
lagged_predictors <- function(n_lags, predictor_df, n_predictors=10){
pred_df <- sapply(1:n_predictors, function(x){
lag(predictor_df[,x],n_lags)
} )
pred_df <- as.data.frame(pred_df)
return(pred_df)
}
# one month lag
predictors_oml <- lagged_predictors(1,predictors)
predictors_oml <- cbind(GDP_LOG, predictors_oml) %>%
drop_na()
# sampling then testing
testIDs <- sample(1:dim(predictors_oml)[1], 0.8*dim(predictors_oml)[1])
training_predictors_oml <- predictors_oml[testIDs,]
testing_predictors_oml <- predictors_oml[-testIDs,]
gdp_oml_lm <- lm(GDP_LOG ~., data = training_predictors_oml)
summary(gdp_oml_lm)
pred_gdp_log_oml <- predict.lm(gdp_oml_lm, newdata = testing_predictors_oml)
ggplot() +
geom_point(aes(x = testing_predictors_oml$GDP_LOG, y = pred_gdp_log_oml)) +
geom_abline(intercept = 0, slope = 1) +
labs(x = "Actual Logged GDP",
y = "Predicted Logged GDP")
training_res <- training_predictors_oml$GDP_LOG - predict(gdp_oml_lm, training_predictors_oml)
holdout_res <- testing_predictors_oml$GDP_LOG - predict(gdp_oml_lm, testing_predictors_oml)
# t-test to check whether predicted residuals are different from zero
t.test(holdout_res, mu = 0)
# test whether holdout residuals and training residuals are the same
t.test(training_res, holdout_res)
```
## Data Processing for Random Forest
Relevant time scale: Monthly
Datasets: Electricity generation, Rigs count, GDP, OECD Business Tendency Survey, Percentage of Civilian Workforce Unemployed >15 weeks
```{r downloading datasets and initial processing, warning=FALSE}
## pct labor force unemloyed for more than 15 weeks
LT_unemployed <- fredr(series_id = "U1RATE", observation_start = as.Date("1973-01-01"))
## yield curve 10year minue 3month
yield_curve <- fredr(series_id = "T10Y3M", observation_start=as.Date('1947-01-01'))
## business tendenc
business_tendenc <- fredr(series_id = "BSCICP03USM665S", observation_start = as.Date("1973-01-01"))
## finding local peaks where m is the number of points on either side of the peak
find_peaks <- function (x, m = 3){
shape <- diff(sign(diff(x, na.pad = FALSE)))
pks <- sapply(which(shape < 0), FUN = function(i){
z <- i - m + 1
z <- ifelse(z > 0, z, 1)
w <- i + m + 1
w <- ifelse(w < length(x), w, length(x))
if(all(x[c(z : i, (i + 2) : w)] <= x[i + 1])) return(i + 1) else return(numeric(0))
})
pks <- unlist(pks)
pks
}
distance_from_peak <- function(current_position, peaks_vec, time_series_value){
if (current_position > peaks_vec[1]){
distance <- peaks_vec-current_position
closest <- min(distance)
distance_pct <- (time_series_value[current_position] - time_series_value[current_position+closest])/time_series_value[current_position+closest]
distance_pct <- ifelse(is.null(distance_pct) == T, 0, distance_pct)
distance_pct
} else { distance_pct <- 0
distance_pct}
}
business_local_peaks <- find_peaks(business_tendenc$value)
distance_peak <- sapply(1:dim(business_tendenc)[1], function(x){
distance_pct <- distance_from_peak(x, business_local_peaks, business_tendenc$value)
distance_pct
})
business_tendencies <- cbind(business_tendenc, distance_peak)
business_tendencies[is.na(business_tendencies)] <- 0
write.csv(business_tendencies, "./data/business_tendencies.csv")
business_sentiment <- business_tendencies
## Electricity Data
# hourly demand
lower48_hourly_demand <- paste("http://api.eia.gov/series/?series_id=EBA.US48-ALL.D.H&api_key=", eia_api_key, sep = "") %>%
fromJSON()
lower48_hourly_demand_list <- lower48_hourly_demand$series %>%
select(data)%>%
.[[1]]
date_time <- lower48_hourly_demand_list[[1]][,1]
demand <- lower48_hourly_demand_list[[1]][,2]
lower48_hourly_demand_df <- cbind(date_time, demand) %>%
data.frame(stringsAsFactors = F) %>%
mutate(date = as_date(ymd(substr(date_time, 1, 8))),
hour = as.numeric(substr(date_time, 10, 11)))
## all generation
# last updated 4/26
all_electricity_generation_monthly <- eia_series(id = "TOTAL.ELETPUS.M")
all_electricity_generation_monthly_df <- unnest(all_electricity_generation_monthly, cols="data")
all_electricity_generation_monthly_df <- all_electricity_generation_monthly_df %>%
select(series_id, units, value, date, year, month)
all_electricity_generation_monthly_df <- all_electricity_generation_monthly_df %>%
select(series_id, value, date, year, month) %>%
mutate(terawatt_value = value/1000,
units = "Terawatthours",
date = as.yearmon(unique(substr(date, 1,7))))
# hourly generation
lower48_hourly_generation <- eia_series(id = "EBA.US48-ALL.NG.H")
lower48_hourly_generation_df <- unnest(lower48_hourly_generation, cols = "data") %>%
select(series_id, units, value, date, year, month)
# monthly generation from hourly
lower48_monthly_generation_df <- lower48_hourly_generation_df %>%
group_by(year, month) %>%
summarise(value = (sum(value))/1000000,
units = "Terawatthours",
date = unique(substr(date, 1,7)))
lower48_monthly_generation_df$date <- as.yearmon(lower48_monthly_generation_df$date)
# intersecting all states with lower 48 | then calculate prop lower48 to extrapolate retroactively
intersecting_months <- inner_join(lower48_monthly_generation_df, all_electricity_generation_monthly_df, by = "date")
colnames(intersecting_months)[3] <- "lower48_terawatt"
intersecting_months$proportion <- intersecting_months$lower48_terawatt/intersecting_months$terawatt_value
lower48_prop <- fivenum(intersecting_months$proportion)[3]
all_electricity_generation_monthly_extrapolated_df <- all_electricity_generation_monthly_df %>%
mutate(lower48_terawatt = terawatt_value*lower48_prop) %>%
filter(date < "Jul 2015") %>%
select(year, month, lower48_terawatt, units, date)
colnames(all_electricity_generation_monthly_extrapolated_df)[3] <- "value"
lower48_monthly_generation_extrapolated_df <- bind_rows(all_electricity_generation_monthly_extrapolated_df, lower48_monthly_generation_df)
write.csv(lower48_monthly_generation_extrapolated_df, "./data/monthly_electricity.csv")
electricity <- lower48_monthly_generation_extrapolated_df
## crude oil and nat gas rigs in operation
total_rigs <- eia_series(id = "TOTAL.OGNRPUS.M")
total_rigs_df <- unnest(total_rigs, cols = "data") %>%
select(series_id, units, value, date, year, month)
total_rigs_df <- total_rigs_df %>%
arrange(date) %>%
mutate(change_first_order = round((value - lag(value, 1))/lag(value,1),4),
change_second_order = ifelse(lag(change_first_order,1) == 0, 0, (change_first_order - lag(change_first_order, 1))/lag(change_first_order,1)))
write.csv(total_rigs_df, "./data/monthly_rigs_count.csv")
rigs <- total_rigs_df
```
```{r data wrangling}
gdp_complete <- gdp %>%
mutate(date = as.yearmon(substr((Date), 1, 7)),
`GDP (Continuously Compounding)` = `GDP (Continuously Compounding)`*100,
`GDP (YoY % Change)` = `GDP (YoY % Change)`*100,
gdp_change_lagged = lag(`GDP (Continuously Compounding)`, 1)) %>%
filter(date >= "Jan 1973") %>%
select(date, `GDP (YoY % Change)`, `GDP (Continuously Compounding)`, gdp_change_lagged)
electricity_complete <- electricity %>%
mutate(date = as.yearmon(date),
rate_generated_change = (value - lag(value,1))/lag(value,1),
rate_generated_change_secondorder = (rate_generated_change - lag(rate_generated_change,1))/lag(rate_generated_change,1)) %>%
select(date, rate_generated_change, rate_generated_change_secondorder)
rigs_complete <- rigs %>%
mutate(date = as.yearmon(substr(date, 1, 7))) %>%
select(date, change_first_order, change_second_order)
colnames(rigs_complete)[2:3] <- c("rigs_change_first_order", "rigs_change_second_order")
business_sentiment <- business_sentiment %>%
mutate(date = as.yearmon(substr(date, 1, 7)),
sentiment_change_first_order = (value - lag(value,1))/lag(value,1)) %>%
select(date, value, distance_peak, sentiment_change_first_order)
colnames(business_sentiment)[2] <- "business_sentiment"
sentiment_period <- rep(0, dim(business_sentiment)[1])
for (i in 2:dim(business_sentiment)[1]){
sentiment_i = business_sentiment$sentiment_change_first_order[i]
#print(sentiment_i)
sentiment_h = business_sentiment$sentiment_change_first_order[i-1]
#print(sentiment_h)
sentiment_h[is.na(sentiment_h)] <- 0
if(sign(sentiment_i) == sign(sentiment_h)){
sentiment_period[i] = sentiment_period[i-1]+1
}
else{
sentiment_period[i] = 1
}
}
business_sentiment$sentiment_period <- sentiment_period
LT_unemployed_complete <- LT_unemployed %>%
mutate(date = as.yearmon(substr(date, 1, 7)),
rate_change_first_order = (value - lag(value,1))/lag(value,1)) %>%
select(date, value, rate_change_first_order)
colnames(LT_unemployed_complete)[2] <- "LT_unemployment"
rate_change_period <- rep(0, dim(LT_unemployed_complete)[1])
threshold_period <- rep(0, dim(LT_unemployed_complete)[1])
for (i in 2:dim(LT_unemployed_complete)[1]){
sentiment_i = LT_unemployed_complete$rate_change_first_order[i]
#print(sentiment_i)
sentiment_h = LT_unemployed_complete$rate_change_first_order[i-1]
#print(sentiment_h)
sentiment_h[is.na(sentiment_h)] <- 0
if(sign(sentiment_i) == sign(sentiment_h)|sign(sentiment_i)==0){
rate_change_period[i] = rate_change_period[i-1]+1
}
else{
rate_change_period[i] = 1
}
if(LT_unemployed_complete$LT_unemployment[i] <= 1.5){
threshold_period[i] = threshold_period[i-1]+1
}
}
LT_unemployed_complete$threshold_period <- threshold_period
LT_unemployed_complete$rate_change_period <- rate_change_period
yield_curve[is.na(yield_curve)] <- 0
yield_curve <- yield_curve %>%
drop_na() %>%
mutate(date = as.yearmon(substr(date, 1, 7)),
value = ifelse(value < 0, -1, 1)) %>%
group_by(date) %>%
summarise(value = min(value)) %>%
mutate(value = as.factor(value))
colnames(yield_curve)[2] <- "curve"
# inner join all together and to get recession dates
econ_situation_nber_df <- inner_join(gdp_complete, electricity_complete, by = "date")
econ_situation_nber_df <- inner_join(econ_situation_nber_df, rigs_complete, by = "date")
econ_situation_nber_df <- inner_join(econ_situation_nber_df, business_sentiment, by = "date")
econ_situation_nber_df <- inner_join(econ_situation_nber_df, LT_unemployed_complete, by = "date")
NBER_recession_months <- econ_situation_nber_df$date %>%
.[c(11:27, 85:90, 103:119, 211:219, 339:347, 420:438)]
econ_situation_nber_df <- inner_join(econ_situation_nber_df, yield_curve, by = "date")
econ_situation_nber_df <- econ_situation_nber_df %>%
mutate(NBER_recession = ifelse(date %in% NBER_recession_months, "Y", "N"),
NBER_recession_next = as.factor(lead(NBER_recession,1)),
NBER_recession_two = as.factor(lead(NBER_recession,2))
) %>%
as.data.frame() %>%
drop_na()
DT::datatable(econ_situation_nber_df)
```
## Random Forest Modeling
Two random forest models were trained. One to predict the probability of a recession the next month, and another for the following month. A previously-trained model was loaded to save time when knitting this document.
```{r random forest onemonth}
# creating test-train data
testIDs <- createDataPartition(econ_situation_nber_df$NBER_recession_next, p = 0.7, list = F)
train_x <- econ_situation_nber_df[testIDs,c(4:17)]
train_y <- econ_situation_nber_df$NBER_recession_next[testIDs]
test_x <- econ_situation_nber_df[-testIDs, c(4:17)]
test_y <- econ_situation_nber_df$NBER_recession_next[-testIDs]
# model tuning
trctrl <- trainControl(method = "repeatedcv",
number = 10, repeats = 5, search = "grid")
mtry <- ncol(train_x)
ntrees <- 101
tunegrid <- expand.grid(.mtry = c(2:mtry))
metric = "Accuracy"
# following can be commented out if loading a pretrained model from file
# parallel processing to speed things up
# random forest model
cl <- makePSOCKcluster(5)
registerDoParallel(cl)
rf_recession <- train(x = train_x, y = train_y, method = "rf", metric = metric, trControl = trctrl, tuneGrid = tunegrid, ntree = ntrees)
stopCluster(cl)
saveRDS(rf_recession, "./models/rf_1mth_all.RDS")
rf_recession <- readRDS("./models/rf_1mth_all.RDS")
predict_y <- predict(rf_recession, newdata = test_x)
predict_y_prob <- predict(rf_recession,newdata = test_x, "prob")
predict_y_prob$date <- econ_situation_nber_df$date[-testIDs]
predict_y_prob$actual <- test_y
predict_y_prob$predicted <- predict_y
cm_1mth_all <- confusionMatrix(predict_y, reference = test_y)
saveRDS(cm_1mth_all, "./models/confusion_matrix_1mth_all.RDS")
cm_1mth_all$overall
prob_model <- ggplotly(ggplot(predict_y_prob, aes(x = date, y = Y)) +
geom_point(aes(color = actual)) +
geom_hline(yintercept = 0.50) +
labs(y = "Probability", x = "Date", colour = "NBER",
title = "Probability of a Recession a Month Ahead", subtitle = "Probabilistic Prediction of a Random Forest Model"))
saveRDS(prob_model, "./models/ggplot_prob_model.RDS")
prob_model
```
```{r random_forest two month, fig.height=8, fig.width=8}
# creating test-train data
testIDs <- createDataPartition(econ_situation_nber_df$NBER_recession_two, p = 0.7, list = F)
train_x <- econ_situation_nber_df[testIDs,c(4:17)]
train_y <- econ_situation_nber_df$NBER_recession_two[testIDs]
test_x <- econ_situation_nber_df[-testIDs, c(4:17)]
test_y <- econ_situation_nber_df$NBER_recession_two[-testIDs]
# model tuning
trctrl <- trainControl(method = "repeatedcv",
number = 10, repeats = 5, search = "grid")
mtry <- ncol(train_x)
ntrees <- 101
tunegrid <- expand.grid(.mtry = c(2:mtry))
metric = "Accuracy"
# following can be commented out if loading a pretrained model from file
# parallel processing to speed things up
# random forest model
cl <- makePSOCKcluster(5)
registerDoParallel(cl)
rf_recession2 <- train(x = train_x, y = train_y, method = "rf", metric = metric, trControl = trctrl, tuneGrid = tunegrid, ntree = ntrees)
stopCluster(cl)
saveRDS(rf_recession2, "./models/rf_2mth_all.RDS")
rf_recession2 <- readRDS("./models/rf_2mth_all.RDS")
predict_y <- predict(rf_recession2, newdata = test_x)
predict_y_prob <- predict(rf_recession2,newdata = test_x, "prob")
predict_y_prob$date <- econ_situation_nber_df$date[-testIDs]
predict_y_prob$actual <- test_y
predict_y_prob$predicted <- predict_y
cm_2mth_all <- confusionMatrix(predict_y, reference = test_y)
saveRDS(cm_2mth_all, "./models/confusion_matrix_2mth_all.RDS")
cm_2mth_all$byClass
prob_model_2 <- ggplotly(ggplot(predict_y_prob, aes(x = date, y = Y)) +
geom_point(aes(color = actual)) +
geom_hline(yintercept = 0.50) +
labs(y = "Probability", x = "Date", colour = "NBER",
title = "Probability of a Recession 2 Months Ahead", subtitle = "Probabilistic Prediction of a Random Forest Model"))
saveRDS(prob_model_2, "./models/ggplot_prob_model_2.RDS")
prob_model_2
```
```{r testing for predicted march data}
gdp_change_lagged <- -0.655704
test_data_2 <- cbind(gdp_change_lagged, electricity_complete[567,2:3], rigs_complete[567,2:3], business_sentiment[567,2:5], LT_unemployed_complete[567, 2:5], yield_curve[459,2])
predict_y_prob_april20 <- predict(rf_recession,newdata = test_data_2, "prob")
predict_y_prob_may20 <- predict(rf_recession2,newdata = test_data_2, "prob")
econ_state <- c("Expansion", "Contraction")
predict_y_prob_april20
predict_y_prob_may20
forecast_df2 <- cbind(econ_state, t(predict_y_prob_april20), t(predict_y_prob_may20)) %>%
data.frame()
colnames(forecast_df2) <- c("Econ. State", "April 2020", "May 2020")
kable(forecast_df2, caption = "R.F. Predictions based on March 2020") %>%
kable_styling(bootstrap_options = c("striped","hover", "condensed"))
```
### Visualizing Random Forest
The following selects a decision tree from the random forest model and plots it. `tree_func` adopted wholesale from [here](https://shiring.github.io/machine_learning/2017/03/16/rf_plot_ggraph)
```{r plotting random forest, warning=FALSE, message=FALSE}
# https://shiring.github.io/machine_learning/2017/03/16/rf_plot_ggraph
tree_func <- function(final_model,
tree_num) {
# get tree by index
tree <- randomForest::getTree(final_model,
k = tree_num,
labelVar = TRUE) %>%
tibble::rownames_to_column() %>%
# make leaf split points to NA, so the 0s won't get plotted
mutate(`split point` = ifelse(is.na(prediction), `split point`, NA))
# prepare data frame for graph
graph_frame <- data.frame(from = rep(tree$rowname, 2),
to = c(tree$`left daughter`, tree$`right daughter`))
# convert to graph and delete the last node that we don't want to plot
graph <- graph_from_data_frame(graph_frame) %>%
delete_vertices("0")
# set node labels
V(graph)$node_label <- gsub("_", " ", as.character(tree$`split var`))
V(graph)$leaf_label <- as.character(tree$prediction)
V(graph)$split <- as.character(round(tree$`split point`, digits = 2))
# plot
plot <- ggraph(graph, 'dendrogram') +
theme_bw() +
geom_edge_link() +
geom_node_point() +
geom_node_text(aes(label = node_label), na.rm = TRUE, repel = TRUE) +
geom_node_label(aes(label = split), vjust = 2.5, na.rm = TRUE, fill = "white") +
geom_node_label(aes(label = leaf_label, fill = leaf_label), na.rm = TRUE,
repel = TRUE, colour = "white", fontface = "bold", show.legend = FALSE) +
theme(panel.grid.minor = element_blank(),
panel.grid.major = element_blank(),
panel.background = element_blank(),
plot.background = element_rect(fill = "white"),
panel.border = element_blank(),
axis.line = element_blank(),
axis.text.x = element_blank(),
axis.text.y = element_blank(),
axis.ticks = element_blank(),
axis.title.x = element_blank(),
axis.title.y = element_blank(),
plot.title = element_text(size = 18))
print(plot)
}
tree_num <- which(rf_recession$finalModel$forest$ndbigtree == min(rf_recession$finalModel$forest$ndbigtree))
tree_plot <- tree_func(final_model = rf_recession$finalModel, tree_num[1])
saveRDS(tree_plot, "./models/tree_plot.RDS")
```