Dheer Toprani 11/18/2021
The Google trends data has been downloaded from trends.google.com, and the csv files have been placed in the rawdata directory. The following tickers were chosen: AMZN, AAPL, TSLA, NVDA. Monthly worldwide trend data for each ticker was collected between the time period of 2011-01-01 to 2020-12-31.
Stock price data was collected using the getSymbols function in the tidyquant package. The data is sourced from Yahoo Finance.
# Setting dates, tickers, etc.
tickers <- c("AMZN", "AAPL", "TSLA", "NVDA")
start_date <- "2010-12-01"
end_date <- "2020-12-31"
e = new.env()
# Read all ticker files simultaneously into "trends" dataframe
myfiles <- paste0("rawdata/", tickers, ".csv")
temp <- lapply(myfiles, read.csv, skip=1)
trends <- temp %>% reduce(inner_join, by = "Month")
# Process trends dataframe
colnames(trends) <- str_remove(colnames(trends), "...Worldwide.") # Fix column names
trends$Month <- paste0(trends$Month, "-01") # Make "Month" column YYYY-mm-dd format
trends[trends == "<1"] <- 1 # Change all "<1" results to 1.
trends[-1] <- trends[-1] %>% mutate_if(is.character, as.integer) # Change all character type columns to integers
trendChange <- trends %>% mutate_at(tickers, function(x) { return ((x-lag(x))/lag(x)) }) # Create a new df to calculate change
colnames(trendChange) <- c("Month", paste0(tickers, ".Change")) # Rename columns of new df
# Fetch monthly price data for all tickers
sapply(tickers, getSymbols, periodicity="monthly", from=start_date, to=end_date, env=e) # Store data in environment e## AMZN AAPL TSLA NVDA
## "AMZN" "AAPL" "TSLA" "NVDA"
After cleaning the data, we combined it into a single dataframe and processed it further. The first few rows of the processed table are shown below.
# Merge data into single table
data <- Reduce(merge, lapply(ls(name=e), get, envir=e))
combined <- as.xts(trendChange[,-1], order.by = as.Date(trendChange[,1])) %>%
merge(data, all=T)
# Calculate Rate of Change for Close prices for each stock ticker
temp <- combined %>%
Cl() %>%
ROC()
colnames(temp) <- paste0(tickers, ".ClROC") # Rename column
combined2 <- merge(combined, temp) %>% na.omit() # Merge to data table and omit NA values
head(combined2)## AMZN.Change AAPL.Change TSLA.Change NVDA.Change AAPL.Open AAPL.High
## 2011-02-01 -0.1666667 -0.24444444 0 -0.20000000 12.18929 13.03214
## 2011-03-01 0.2000000 -0.02941176 0 0.00000000 12.69536 12.91679
## 2011-04-01 0.1666667 0.00000000 0 -0.25000000 12.53964 12.68321
## 2011-05-01 -0.1428571 -0.36363636 0 0.00000000 12.49071 12.56536
## 2011-06-01 -0.3333333 0.23809524 0 -0.08333333 12.45964 12.57607
## 2011-07-01 0.5000000 0.46153846 0 -0.18181818 11.99821 14.44643
## AAPL.Low AAPL.Close AAPL.Volume AAPL.Adjusted AMZN.Open AMZN.High
## 2011-02-01 12.06143 12.61464 9295949600 10.81535 170.52 191.40
## 2011-03-01 11.65214 12.44679 11306458800 10.67144 173.53 181.57
## 2011-04-01 11.43429 12.50464 9253829200 10.72105 181.58 197.80
## 2011-05-01 11.76500 12.42250 6912060400 10.65062 196.57 206.39
## 2011-06-01 11.08929 11.98821 9263850400 10.27828 196.06 206.25
## 2011-07-01 11.93571 13.94571 10653946800 11.95657 205.55 227.20
## AMZN.Low AMZN.Close AMZN.Volume AMZN.Adjusted NVDA.Open NVDA.High
## 2011-02-01 169.51 173.29 95776400 173.29 6.0325 6.5425
## 2011-03-01 160.59 180.13 118979100 180.13 5.6750 5.7475
## 2011-04-01 175.37 195.81 116749400 195.81 4.6750 5.1100
## 2011-05-01 190.88 196.69 106274500 196.69 5.1050 5.1300
## 2011-06-01 181.59 204.49 95563700 204.49 4.9875 4.9950
## 2011-07-01 203.61 222.52 92808500 222.52 3.9850 4.0700
## NVDA.Low NVDA.Close NVDA.Volume NVDA.Adjusted TSLA.Open TSLA.High
## 2011-02-01 5.4450 5.6650 2402576400 5.203927 4.862 5.098
## 2011-03-01 4.2525 4.6150 2759758000 4.239387 4.810 5.742
## 2011-04-01 4.2075 5.0000 1532207200 4.593053 5.490 5.636
## 2011-05-01 4.2800 5.0100 1716251600 4.602238 5.520 6.056
## 2011-06-01 3.7150 3.9850 1595378400 3.660663 6.000 6.300
## 2011-07-01 3.3975 3.4575 1455165600 3.176096 5.814 6.088
## TSLA.Low TSLA.Close TSLA.Volume TSLA.Adjusted AMZN.ClROC
## 2011-02-01 4.222 4.778 146517500 4.778 0.040119213
## 2011-03-01 4.354 5.550 162816500 5.550 -0.013395845
## 2011-04-01 4.840 5.520 149474500 5.520 0.004637578
## 2011-05-01 5.104 6.028 151744000 6.028 -0.006590671
## 2011-06-01 5.100 5.826 198636500 5.826 -0.035585344
## 2011-07-01 5.326 5.634 105510500 5.634 0.151248220
## AAPL.ClROC TSLA.ClROC NVDA.ClROC
## 2011-02-01 0.021287913 -0.054113673 -0.008751879
## 2011-03-01 0.038712389 -0.204995027 0.149775879
## 2011-04-01 0.083465951 0.080126044 -0.005420067
## 2011-05-01 0.004484104 0.001998003 0.088037420
## 2011-06-01 0.038890204 -0.228898603 -0.034084622
## 2011-07-01 0.084497904 -0.141991528 -0.033510989
After the data was sufficiently processed, we formulated our trading strategy described below. We normalized the month-over-month change in trend data to create monthly portfolio weights that sum up to 1.
# Select column names with the ".Change" suffix
ChangeSuffix <- colnames(combined2)[colnames(combined2) %>% endsWith(".Change")]
# Strategy 2: Normalize ".Change" columns to sum to 1, and allocate according to normalized weights
StrategyRule2 <- function(df, tickers) {
ChangeSuffix <- paste0(tickers, ".Change") # Fetch all MoM change in "Interest over time" trend data from Google
normalizer <- 1/rowSums(df[,ChangeSuffix]) # Find a normalizing coefficient so that the portfolio weight will add up to 1
temp <- df %>% mutate_at(ChangeSuffix, function(x) { x*normalizer }) # Multiply the MoM change in trends with the normalizer coefficient
# Change column names, and merge with the input dataframe
colnames(temp) <- paste0(tickers, ".NormSignal") %>% na.omit()
return(cbind(df, temp[,1:length(tickers)]))
}
combined3 <- StrategyRule2(as.data.frame(combined2), tickers)Once the trading signal and strategy was established, we calculated our returns by multiplying portfolio weights with corresponding returns.
# Find Signal and ClRoc columns
SignalSuffix <- colnames(combined3)[colnames(combined3) %>% endsWith(".NormSignal")]
ClRocSuffix <- colnames(combined3)[colnames(combined3) %>% endsWith(".ClROC")]
# Multiply the values and merge the new columns with the combined dataset
temp <- combined3[,SignalSuffix]*combined3[,ClRocSuffix]
colnames(temp) <- paste0(tickers, ".Multiple")
combined4 <- cbind(temp, combined3)We also created benchmark passive portfolios to compare our strategy against. One benchmark portfolio represented all the stocks weighted equally over the 10-year period. The other “Weighted Benchmark” represents the average portfolio weights of our trading strategy (shown in the table below) held passively over the 10-year period.
# Add portfolio multiple, average benchmark, and weighted benchmark to the combined dataset
avgPortfolioWeights <- as.matrix(colMeans(combined4[,paste0(tickers, ".NormSignal")]))
temp <- cbind(
rowSums(combined4[,paste0(tickers, ".Multiple")]),
rowMeans(combined4[,ClRocSuffix]),
rowSums(avgPortfolioWeights*combined4[,ClRocSuffix])
)
colnames(temp) <- c("Portfolio.Multiple", "Benchmark.AvgClROC", "Benchmark.WeightedClROC")
combined5 <- cbind(combined4, temp)
# Format and display Weighted Benchmark weights
colnames(avgPortfolioWeights) <- "Weights"
rownames(avgPortfolioWeights) <- paste0(tickers, ".WeightedBenchmarkWeights")
avgPortfolioWeights## Weights
## AMZN.WeightedBenchmarkWeights 0.3779985
## AAPL.WeightedBenchmarkWeights 0.2864872
## TSLA.WeightedBenchmarkWeights 0.1455330
## NVDA.WeightedBenchmarkWeights 0.1899813
Finally, we compared the performance of our strategy to our benchmarks using various tables and charts.
perf <- combined5 %>% subset(select = c(Portfolio.Multiple, Benchmark.AvgClROC, Benchmark.WeightedClROC))
colnames(perf) <- c("Google Trends-based Strategy", "Benchmark", "Weighted Benchmark")
perf <- xts(perf, order.by = as.Date(rownames(perf)))
table.DownsideRisk(perf)## Google Trends-based Strategy Benchmark
## Semi Deviation 0.1207 0.0513
## Gain Deviation 0.2063 0.0535
## Loss Deviation 0.1432 0.0421
## Downside Deviation (MAR=10%) 0.1062 0.0408
## Downside Deviation (Rf=0%) 0.1032 0.0369
## Downside Deviation (0%) 0.1032 0.0369
## Maximum Drawdown 0.8406 0.3566
## Historical VaR (95%) -0.1599 -0.0848
## Historical ES (95%) -0.3698 -0.1224
## Modified VaR (95%) -0.0256 -0.0896
## Modified ES (95%) -0.4571 -0.1222
## Weighted Benchmark
## Semi Deviation 0.0520
## Gain Deviation 0.0541
## Loss Deviation 0.0453
## Downside Deviation (MAR=10%) 0.0416
## Downside Deviation (Rf=0%) 0.0379
## Downside Deviation (0%) 0.0379
## Maximum Drawdown 0.3561
## Historical VaR (95%) -0.0806
## Historical ES (95%) -0.1296
## Modified VaR (95%) -0.0900
## Modified ES (95%) -0.1296
table.Drawdowns(perf)## Warning in table.Drawdowns(perf): Only 2 available in the data.
## From Trough To Depth Length To Trough Recovery
## 1 2011-03-01 2012-10-01 2019-12-01 -0.8406 106 20 86
## 2 2020-09-01 2020-10-01 2020-12-01 -0.1190 4 2 2
table.Stats(perf)## Google Trends-based Strategy Benchmark Weighted Benchmark
## Observations 119.0000 119.0000 119.0000
## NAs 0.0000 0.0000 0.0000
## Minimum -0.8285 -0.1905 -0.2169
## Quartile 1 -0.0332 -0.0146 -0.0153
## Median 0.0311 0.0260 0.0252
## Arithmetic Mean 0.0439 0.0282 0.0293
## Geometric Mean 0.0216 0.0255 0.0266
## Quartile 3 0.0893 0.0727 0.0712
## Maximum 1.5160 0.2667 0.2502
## SE Mean 0.0196 0.0068 0.0069
## LCL Mean (0.95) 0.0051 0.0147 0.0157
## UCL Mean (0.95) 0.0827 0.0417 0.0430
## Variance 0.0457 0.0055 0.0056
## Stdev 0.2139 0.0745 0.0751
## Skewness 2.6707 0.1567 0.0947
## Kurtosis 21.0889 0.5699 1.0869
charts.PerformanceSummary(perf)
chart.RelativePerformance(perf[,1], perf[,2])
chart.RiskReturnScatter(perf, add.sharpe = c(1), xlim = c(0.1, 1.1))