Saturday, March 19, 2011

Cointegration or Correlation

Does Cointegration works for short term? After several backtests, it seems correlation works better in the shorter time frame.

Testing for Cointegration

Timeframe used: 5 mins bar chart (Over 2 months)
Steps:
1. Check for stochastic trend in the time series. Meaning, I(1) is stationary (TS is stationary after 1st differencing
2. Regress the 2 Time series, do a ACF test on the residuals.
3. A small P-Value => "Cointegration" (For hardcore stats people, "Do not accept Null Hypothesis"

No comments:

Post a Comment