I have a strategy that uses 1-minute price data on several stocks. Not every stock trades during every minute. Missing prices can be filled in using the
(1) prior available price
(2) next available price
(3) the average of (1) and (2)
(4) a regression model using the returns of other stocks that did trade
I think (4) is best, but it takes more programming, modeling (which stocks are used to predict a given stock?), and CPU time. Price data in a backtest is used to both generate signals and to set transaction prices. Even if I interpolate missing data to create the time series used in indicators, I may label those prices and avoid having the backtest trade during those periods.
How do people deal with missing intraday data? One can program various approaches and see if they produce substantially different results.
Submitted November 12, 2020 at 06:10PM by Beliavsky