Lead/Lag Analysis

doubleslit · October 7th, 2009, 11:52 pm

What is the best method for lead/lag analysis of 2 time series? Cross-correlation? Cointegration?

Aaron · October 8th, 2009, 1:03 am

Cointegration is more general but I wouldn't call it a "method" so much as an approach. Correlation-based analysis can be done using more straightforward recipes.A good way to start is with visualization. Look at the series, both in time series and cross-sectional scatter plots at various leads and lags.The application also matters.

quantmeh · October 8th, 2009, 3:09 pm

"granger causality" could be relevant too

beaker · October 8th, 2009, 6:29 pm

Visualization is the best way to start (though is highly underrated). This is true not inly in time series, but anytime you are dealing with data.If you see something in your scatterplots then move to vector autoregressions and tests of granger causality. There are a couple of test statistics for granger causality which you can find in any time series text (or at least citations to them). Note: the results of granger causality tests are sensitive to lag length and how you deal with nonstationarity.

doubleslit · October 13th, 2009, 7:19 pm

DELETED

quantmeh · October 14th, 2009, 12:14 am

QuoteOriginally posted by: doubleslitDELETEDsmart move there's no point in correlation if the two series are non-stationery

doubleslit · October 14th, 2009, 4:11 pm

QuoteOriginally posted by: jawabeanQuoteOriginally posted by: doubleslitDELETEDsmart move there's no point in correlation if the two series are non-stationeryYah I knew something wasn't right with my analysis. We need joint stationarity for correlation analysis. So if I'm looking at a price to sales ratio series against market returns, do I need to difference the price to sales ratio series first before I do correlation analysis? Or should I explore cointegration between the price to sales ratio series with the market's closing prices?

quantmeh · October 14th, 2009, 4:55 pm

QuoteOriginally posted by: doubleslitSo if I'm looking at a price to sales ratio series against market returns, do I need to difference the price to sales ratio series first before I do correlation analysis? Or should I explore cointegration between the price to sales ratio series with the market's closing prices?it depends. i'd test for cointegration, look at johanssen (whatever the spelling is) test. if there's a reason to believe in cointegration, then do it. otherwise, diff would be an option. i dont think there's one easy rule to decide.