Delisted stock historical data from www.csidata.com
Author: tedclimo
Creation Date: 12/3/2010 4:43 PM
profile picture

tedclimo

#1
I'm currently researching impact of delisted stocks(survivorship bias) on various models.
CSI says they can provide historical data going back 20 yrs(including delisted stocks for a mere $1000(proud of their data, it seems)
They say data can be in in following formats(proprietary CSI, Metastock, Ascii or Excel).
My ? is this, does anyone have experience with CSI data imported into WLP 6.0?

Before I bite the $1000 bullet, to insure survivorship bias is not an issue in various models, I would greatly appreciate all insights from our community as to the benefits/problems of incorporating CSI's data.

Your comments will be greatly appreciated.
T!

profile picture

Roland

#2

tedclimo,

A lot of studies have been done on survivorship bias. One I recall stated that it would reduce your long term performance by about 3%. So it is not that much relevant. What is, however, is your method of play. If you use a multiple position dip-buyer strategy, you might fall on a suspended stock that never re-opens (I have in mind the REFCO debacle for instance).

So my advice is: keep your money. That data is not worth it. You might even be able to find it free somewhere.

Regards

profile picture

tedclimo

#3
To Roland,
Free would be nice, but not holding my breath.
REFCO is great example of how not to play the dip-buying game. My decades of autopsies(on failed trades) have taught me to build filters, so I have an insurance policy, so to speak... I incorporate Fundamental filters that assure stock is relatively immune from bankruptcy, and if delisted, would most likely be due to takeover.
Here's the best news yet, I have an acquiantence at high profile firm who has done numerous proprietary studies on Survivorship bias. He affirms that with appropriate fundamental filters in place. Stats are actually better with delisted stocks included, primarily b/c takeovers generally result in price pops.

To everyone else,
Any info/experience you can share on CSI as a datasource in conjunction w WLP 6.0 would be greatly appreciated.
profile picture

Roland

#4

tedclimo,

I see your point. However, be assured there is no appropriate fundamental filter to protect you from sudden suspended stocks that are to be declared bankrupt. REFCO was not the only one in recent years. Remember that when Enron went under, the first analyst to recommend a sell did it on his “fundamentals” and at 0.69 cents a share. Or that after 120 years in business, Lehman made a historic high only to go bankrupt in just 9 months with a “fundamental” buy rating all the way down.

Including bankrupt companies in your strategies is fine; it will help design safety measures to protect your capital. But as I mentioned before a reduction of 3% long term might seem at first glance trivial until you realize that the average long term return is around 10%. This does mean that you will lose some 30% of your potential performance. And then it is not so trivial.

However, there is a very simple remedy to the survival bias dilemma and that is never trade stocks under $10.00. This simple rule will not only eliminate most potential going bankrupt companies, but it will also keep most of the potential acquirable companies in your watch list.

Regards

profile picture

Eugene

#5
Hi Ted,

Just a note that the cost for their data might not be optimal. Premiumdata.net offers delisted securities data for just a fraction of the CSI's price.
profile picture

tedclimo

#6
Eugene,
Thanks for tip, For $307, I get all US listed stocks(including delisted)going back to 1985. 25yrs is more than enough to validate impact of delisted stocks on various models. For others interested, they also have stock history(including delisted) going back to 1950(for another $400).
Bad news is... no Fundamental data. So fundamental filters impossible on delisted securities.

Roland,
Thanks for sharing your insights. Once I get the data(w delisted included), I'll test several public models & share the results with the community on this thread. If you have a particular script you would like to see a comparison on, please let me know here or privately at tedclimo@bellsouth.net.

All the best to you both,
T!
profile picture

Roland

#7

tedclimo,

You could save yourself a lot of work by first consulting Google scholar where even the first listing of 16 000 would answer most of your questions free as many other people have done the same research.

Try this one where the conclusion is about 3% as stated previously.

Regards and good luck.

profile picture

tedclimo

#8
Roland,
Your wisdom is greatly appreciated. Thanks for sharing. These academician white papers do generally agree that delisted stocks have a negative impact on studies they have done. Your solution of trading only $10+ stocks does remove most risk of bankruptcy.

Color me contrarian, but there's more than one way to skin a cat(or so the old saying goes). The acquaintence I referred to above is a trader and published trading model researcher at a high profile firm. He assures me that their proprietary research disputes what the academics publish and that delisted stocks are actually a boost to historical performance.

So there lies the dilemma; do I believe a professional trader or do I believe an academic(who may or may not have trading experience).

I have been building trading models for private investors for over a decade. Successful traders build in robust filters. I have not seen a white paper yet that claims to do so. So the wisest approach seems to be to do the survivorship bias studies myself with private robust trading models. Then I will know with certainty whether the academic's are correct or not.... as it applies to my techniques.

Again, thanks for your wisdom.

Will let you know what I find.
profile picture

Roland

#9

Ted,

All our trading research can be resumed in a single equation comprised of only two variables; and these are the number of shares held over a price differential. The best expression for this was provided by Schachermayer with his payoff matrix formula:



where H stands for the number of shares held and delta S for the price differential between purchase and sale. The formula translates to the cumulative sum of all profits generated by your trading strategy. Any trading method can be expressed using Schachermayer’s formula.

You can select the price variations you want over any time interval and here hindsight makes us believe that what we have selected in the past should be the same “in all probabilities” in the future. But that is not very accurate. There is nothing we can do to change the future price variations of any of the stocks we trade. And therefore, we are bound long term to achieve no better than the secular price trend or close to it. This is about 10% return long term. Take 3% off for survivorship bias, another 1% for commissions, slippage and errors, 3% for inflation and 3.5% for not reinvesting dividends. This would result in a net long term negative 0.5% for playing the game. And then we wonder why over 80% of traders lose trying to outperform the market.

Are there ways to outperform you ask? Well yes and that is by controlling both parameters in the equation. First by making better price selections (choosing better stocks, stocks that have a future) but this will not change the nature of price variations; and second by controlling the inventory on hand (the H in the equation) where you have total control. And therefore it all boils down on how you play the game. What is your position sizing algorithm used to maximize your profit scenario? And that is where I think you should concentrate your efforts: in designing better holding matrixes.

Fundamental data can help in the stock selection process but very little in trading. Most fundamental data series are provided late (like one, two or three months late); but not only that, they are subject to revisions the very next month. There is no way in back testing on fundamental data to know if the real data was really there at the time of decision making. And this raises the problem of peeking, curve fitting and over optimizing when back testing. All of which will torpedo your future results.

In my own research, I have read not tens but hundreds of research papers on every facet of the game in order to better understand the nature of the game. Some of it is honest research and some of it is simple data manipulation to make a point which in real life would destroy any portfolio. That is why you should only believe in myself, oops, yourself.

Regards and good luck.