- Image via Wikipedia
Market Monk has been doing an interesting series on Survivorship Bias – and how much of a problem it can be. What shocks me is that not one developer of system testing software has dealt with this. Here’s the deal – if you’re testing on the Nasdaq 100, your test is likely to be invalid because of the stocks that have gone away. Or if you’re creating an index based on the S&P 500, then that index will likely not be valid going back in time, say, a year before the index has problems.
So what’s the answer here? One possible answer – create a custom index (with all components) that includes all stocks that have traded in the index. Now, you can get a list and data for all the delisted stocks – but the issue is creating an index that either includes all of those delisted stocks, or have the list be dynamic – meaning that it updates as changes are made. As of this writing, I don’t think one data vendor or system software creator has a tool to deal with the issue. I’ve been chatting with the folks at Norgate Premimum Data services – it sounds like they might be working on something to deal with the issue later this year.
![Reblog this post [with Zemanta]](http://img.zemanta.com/reblog_e.png?x-id=8848677c-6b1f-46cb-ba93-dfe226628f81)
Hi Damian,
I also have been chatting with Norgate as well as a lot of other folks. My interest is in creating “true” custom indicators on the QQQQs since it’s inception and as you know it has had many changes/substitutions over the years. A lot of those stocks that use to be in the index are no longer trading.
While the price data is available (for a good chunk of $), there is no easy method to create proper analysis automatically with software.
I have a few ideas and if interested shoot me an email.
MM
dskills,
Isn’t this the point of trading the index versus trading the issues themselves?
ie if your system takes the Nasdaq 100 components and then trades them, you have serious worries about survivorship bias. However if your system trades the Nasdaq 100 index then while which companies influence the index movements might change overtime you are being auto-rebalanced.
This auto-rebalancing is one of the reasons things like the s&p500 outperform over time of course (and so people investing in an index are more ‘active’ than they might realise).
All in all, while the rebalancing of the indexes might itself represent a move to live on with the survivors, trading systems trading the indices essentially dont have to worry about the issue.
Or is there something I’m missing?
–Q
Yeah Quarrel – but I might want to create a system that trades individual stocks drawn from the Nasdaq 100 vs. trading the index. Or, more commonly, I want to create an indicator based on an index – a simple example would be the % of stocks above/below a 50-day average. I need accurate components to have an accurate indicator. Then I might create a system that trades the index itself based on that. Hope that explains my thinking – thanks for the comment!
dskills,
ah- thank you. Yes, for computing breadth indicators I can see that it would get tough.
I missed the distinction, but yes, anything that may even be trading the index, but relies on examining the individual constituents does start to get tricky pretty quickly…
–Q
Not at all Q – thanks for the comment and for stopping by….
http://wl4.wealth-lab.com/cgi-bin/WealthLab.DLL/getpage?page=IndexLab.htm
I know they’ve moved on to 5.0 but you can still dl this plugin.
The historical record shows that most stocks under-perform the indexes they are members of and many of them lose virtually all their value. Survivorship bias is VERY significant:
http://michaelcovel.com/pdfs/TrendingStocksDriveTheMarket.pdf
Agreed Cyrus – I just wish I could find a good solution to the problem!