Hello, I am trying to compute a 20x20 correlation matrix of 20 equities index. However, the trading days of the markets are not the same, e.g. the markets in the states are closed in Xmas but the ones in Asia are open. The 20 indexes have different trading calendars. What is a good way to compute the correlation matrix of the 20 stock markets? Thank yo

The riskmetrics technical guides have details on filling in missing data, from memory it's in their section on volatility and correlation

QuoteOriginally posted by: smalldoorboyHello, I am trying to compute a 20x20 correlation matrix of 20 equities index. However, the trading days of the markets are not the same, e.g. the markets in the states are closed in Xmas but the ones in Asia are open. The 20 indexes have different trading calendars. What is a good way to compute the correlation matrix of the 20 stock markets? Thank yoUse EM algorithm. You might even find matlab code already written.

There are a number of algorithms to estimate the covariance matrix. The simplest one is to use overlapping three-day returns.

http://www.mathworks.com/matlabcentral/ ... hors/20228 have to be interesting in the EM case, not that I have used it, but I guess I probably will now. That is kind of the point of keeping a small code library (with you).

QuoteOriginally posted by: cryptic26QuoteOriginally posted by: smalldoorboyHello, I am trying to compute a 20x20 correlation matrix of 20 equities index. However, the trading days of the markets are not the same, e.g. the markets in the states are closed in Xmas but the ones in Asia are open. The 20 indexes have different trading calendars. What is a good way to compute the correlation matrix of the 20 stock markets? Thank yoUse EM algorithm. You might even find matlab code already written.Thanks.When I use EM algorithm, I face a problem. Let's say I have a missing closing price data P(t) on day t. It brings me two missing returns r(t) and r(t+1). When I use EM algorithm, it gives me the two missing returns r(t)* and r(t+1)*, but I use r(t)* to get P(t)* and use P(t)* and P(t+1) to get r(t+1)** in order to match the closing price P(t+1). Therefore r(t+1)** will be different from r(t+1)* and the expected log likelihood won't be maximized as the case of r(t)* and r(t+1)*.Is my method of matching P(t+1) not working with EM algorithm?