Embedded Files
MEASUREMENTS
MEASUREMENTS
We use mean-square error to measure the performance of the results
MODEL SELECTION
MODEL SELECTION
Random Forest Model
Naive Arima Model
FB Prophet Model
FB Prophet Model:
The graph shows that the random forest model can predict well in a short period, but on average, FB prophet model gives better S&P500 index prediction and smaller mean-square error.
FEATURES
FEATURES
After AIC feature selection, we consider the following combination as our features.
10 Year U.S. Treasury Bond Yield Rates (tby)
Federal Funding Rate (ffr)
Federal Total Assets (fta)
Earning-Per-Share of S&P 500 (eps)
Dividend Yield of S&P 500 (div)
Unemployment Rate (une)
West Texas Intermediate Oil Index (wti)
Producer Price Index (ppi)
Retail and Food Services Sales (rfs)
FB_prophet_predicted_price (fbsp) times tby
fbsp times ffr
fbsp times div
eps times tby
eps times ffr
eps times div
TRAINING
TRAINING
We use linear regression model on these features to optimize the difference between S&P500 close price and predicted price by FB prophet model.
RESULTS
RESULTS
BEFORE COVID:
BEFORE COVID:
Mean Square Error is improved by ~22%
INCLUDE COVID
INCLUDE COVID
Even though including covid as our test period produces worse mean square error, but our improved model predicts a big drawdown and the recovery during the Covid time.
Page updated
Google Sites
Report abuse