Traders' Roundtable

I just wanted to get thoughts from the forum on the idea of excluding data from the stock market bubble/pop years (99-01) for backtesting.

Considering the infrequency of such bubbles in a single market, would it be valid to exclude test results from this period, or at least consider them seperately?

Another issue that I have is with the fact that the long-term bull market in US stocks began in 1980, and it's difficult to get data for individual stocks pre-dating this. I don't really feel comefortable that a test system designed exclusively around a long-term bull market will be valid going forward. On the other hand, even if I could get stock data from the '70s or earlier, would it be valid considering the structural changes to the markets over the past 20 years (ie. the internationalization, explosion of volume, etc.)?

I guess the central question is, how does one decide what data is most likely to be representative going forward? Perhaps an impossible question to answer.

Bubbles are just trends and a good trading system should be able to take advantage. Excluding bubbles you are just curfitting your system. Even Nobel prizes went broke forgetting about extreme events.
A good trend following system is aimed to take advantage of buubles.
And what about diversification? Markets are not just stocks, you can trade interest rates, currencies,energy,metals, grains and meats.
Trend following systems trade many markets simultaneously to reduce the overall risk and have the chance to ride bubbles or trends whenever they appear. Markets are markets, they never change. The only thing is sure about market is that prices changes in trends.

kianti,
I see your point that a good trend following system should be built to take advantage of bubbles, but I also think that it's important to consider the results without that bubble period included. It's certainly only a guess on my part, but I doubt that a similar scenario will occure again in US equity markets for a very long time.

c.f. made a point in another post that stock have an upward bias, and that was certainly true through the '80s and '90s, but if you look further back at the DJIA through the '30s and '40s, there didn't appear to be any kind of an upward bias through that timeframe. Looking at Japan, the Nikki ran up all through the '80s and ended in a bubble, but has been on a slow decline since. I know this is anecdotal evidence, but I think it's worth considering that the stock market will trade considerably differently now that the party is over. So I'm not so willing to assume that what worked well in the stock market over the past 20 years will work over the next 20.

Perhaps it would be worthwhile to test a system against the Japanese market data over the past 10 years. But I'm not sure how to access Japanese historical data, nor if there are major differences in the way stocks trade there.

I totally agree with you in regards to diversification, but I have less than US$10k available for trading, so it would be difficult to trade futures. I am considering currencies, but first I need to learn more about how that market works.

Jason

For a 10k trader, there are some alternatives albeit none too pretty.

Many mini futures contracts have been introduced recently, have a look on the sites of CME,CBOT and NYBOT.
Diversification, as far as I know, is essential in any trend following system. You need to spread your bets on uncorrelated market as much as you can, taking small losses and waiting for the big wave.
No diversification -> No money mangement -> No trend following system.

Kianti,

I've seen that claim before:

At the end of the day, trend-following is just replicating a diversified portfolio of call options.

Its not quite right. The big difference is that with the options you can never lose more than wht you paid for them whereas with the futures you can lose every cent in your account. Imagine being short every currency and gold when a major event seriously damaged the US. Its possible that you might be saved by exchange closure but maybe not

This is the type of event that no historical testing can cover.

John

Kiwi,

I agree with you that there is more risk in futures compared to the options, regarding the stops or premiums paid; that's the reason I used the "replication" term; replicating works well on liquid markets; in case of crashes, extreme events, fat tails or whatever the strategy could be very painful, see '87 crash and portfolio insurance.
On the other side the premium you pay to buy calls could be very expensive and not justify a trade from a risk/reward point view.

Agreed

I think the idea of not expecting the next few years of stock trading to be as good at the "bubble years" is a valid one.

However, I wouldn't exclude it from back-testing. I just might not expect results to be as great as those years might suggest.

I generally test stocks using at least 20 years of data. This give two crashes, two bull markets and many years of chop. Any system that works well during all these periods will hold up pretty well.

At least that's the theory.

The stock markets have an upward bias even when not considering the "bubble years". Economies grow so it is not unreasonable to expect that stock prices will go up as well. It may not be reasonable to expect them to grow faster than the economy, however.