tag:blogger.com,1999:blog-2308881368473451342.post6600358451461112084..comments2023-08-03T03:50:28.359-04:00Comments on The Mousetrap: 03/30/2014 End of Quarter reportteclontzhttp://www.blogger.com/profile/08397130086867288340noreply@blogger.comBlogger6125tag:blogger.com,1999:blog-2308881368473451342.post-56150418719873260752014-04-02T10:04:49.784-04:002014-04-02T10:04:49.784-04:00There's also the problem of constituency itsel...There's also the problem of constituency itself (which you just mentioned). I tried to get the monthly histories from SDRS, but their histories were not that granular, and (which is worse) were corrupted. They sometimes had stocks in the wrong sectors entirely. When I tried to get corrected data, they were unable to do so.<br /><br />The nearest I could get for the backtest was a reconstruction I had to piece together myself for which of the presently existing stocks were most likely to have been in or near the top holdings for any given time period.<br /><br />After weeks of work I got something that was good enough for a sanity check, but in no way could be considered a pristine reproduction of what would have occurred.<br /><br />The good news, though, is that the worse the data was, the worse the returns were. The OPPOSITE would have been a problem. I'd much rather under promise and over deliver than over promise and under deliver.teclontzhttps://www.blogger.com/profile/08397130086867288340noreply@blogger.comtag:blogger.com,1999:blog-2308881368473451342.post-19560021906552214512014-04-02T09:56:54.926-04:002014-04-02T09:56:54.926-04:00Ditto, Compustat Xpressfeed which was rather expen...Ditto, Compustat Xpressfeed which was rather expensive but data accuracy was good. The market data was EOD only, so it's inadequate for intraday trading signals, unless you only use it for constituency info and source the intraday data from another feed? At least for S&P 500 equities, it's not like companies "disappeared" but merely left the index, etc.Doron Shermanhttps://www.blogger.com/profile/10909796071792579259noreply@blogger.comtag:blogger.com,1999:blog-2308881368473451342.post-69936417220885640312014-04-01T15:23:20.550-04:002014-04-01T15:23:20.550-04:00True. I've accessed Compustat in the past. F...True. I've accessed Compustat in the past. For now I'm content to use it as a benchmark to be beaten, especially since I've been using earlier drafts of this model live for the past three years.teclontzhttps://www.blogger.com/profile/08397130086867288340noreply@blogger.comtag:blogger.com,1999:blog-2308881368473451342.post-92128096189477462282014-03-31T19:31:45.648-04:002014-03-31T19:31:45.648-04:00There are survivorship-bias free databases, but th...There are survivorship-bias free databases, but they aren't cheap. What timeframe accuracy do you need for market data?Doron Shermanhttps://www.blogger.com/profile/10909796071792579259noreply@blogger.comtag:blogger.com,1999:blog-2308881368473451342.post-56475722427437141362014-03-31T13:38:45.022-04:002014-03-31T13:38:45.022-04:00Probably a little of both. The caveat is that mea...Probably a little of both. The caveat is that mean reversion works both ways.<br /><br />But the backtest actually performs worse than real time experience because ETF constituency data gets less accurate the further you go back (some of the former holdings no longer exist, for instance).<br /><br />teclontzhttps://www.blogger.com/profile/08397130086867288340noreply@blogger.comtag:blogger.com,1999:blog-2308881368473451342.post-84792919538970948562014-03-31T13:16:18.071-04:002014-03-31T13:16:18.071-04:00Tim, the sector model is indeed performing ridicul...Tim, the sector model is indeed performing ridiculously lately (here, I just jinxed it, didn't I). Would you attribute most of that to the mean-reverting concept or to the machine learning optimization?Doron Shermanhttps://www.blogger.com/profile/10909796071792579259noreply@blogger.com