Logo Icon

Time Series Cross-validation 4: Forecasting the S&p 500

Note: This post is NOT financial advice! This is just a fun way to explore some of the capabilities R has for importing and manipulating data.

I finally got around to publishing my time series cross-validation package to github, and I plan to push it out to CRAN shortly:

devtools::install_github("zachmayer/cv.ts")

Then run the following script to check it out:

library(forecast)
library(cv.ts)
set.seed(42)

#Download S&P 500 data and adjust from splits/dividends
library(quantmod)
getSymbols('^GSPC', from='1990-01-01')
#> [1] "GSPC"
GSPC <- adjustOHLC(GSPC, symbol.name='^GSPC')

#Calculate monthly returns
GSPC <- to.monthly(GSPC, indexAt='lastof')
GSPC <- Cl(GSPC)

#Convert from xts to ts
GSPC <- ts(GSPC, start=c(1990,1), frequency=12)

#Define cross validation parameters
myControl <- tseriesControl(
                  minObs=60,
                  stepSize=1,
                  maxHorizon=12,
                  fixedWindow=TRUE,
                  preProcess=FALSE,
                  ppMethod='guerrero',
                  summaryFunc=tsSummary
                  )

#Forecast using several models
result_naive <- cv.ts(GSPC, naiveForecast, myControl, progress=FALSE)
myControl$preProcess <- TRUE
result_autoarima <- cv.ts(GSPC, auto.arimaForecast, myControl, ic='bic', progress=FALSE)
result_ets <- cv.ts(GSPC, etsForecast, myControl, ic='bic', progress=FALSE)
library(reshape2)
library(ggplot2)
plotData <- data.frame(
  horizon=1:12
  ,naive=result_naive$results$MAPE[1:12]
  ,arima=result_autoarima$results$MAPE[1:12]
  ,ets=result_ets$results$MAPE[1:12]
  )
plotData <- melt(plotData, id.vars='horizon', value.name='MAPE', variable.name='model')
print(ggplot(plotData, aes(horizon, MAPE, color=model)) + geom_line())
A line plot comparing the Mean Absolute Percentage Error (MAPE) across different forecast horizons for three models: naive, ARIMA, and ETS. The plot shows that MAPE increases as the forecast horizon extends, with the ARIMA model having the highest error, followed by ETS and then the naive model, which consistently performs the best across all horizons.

Forecasting equities prices is hard!

stay in touch