Ideas u consider or learn when testing.

Discussions about the testing and simulation of mechanical trading systems using historical data and other methods. Trading Blox Customers should post Trading Blox specific questions in the Customer Support forum.
Post Reply
Nathan
Roundtable Knight
Roundtable Knight
Posts: 127
Joined: Sun Jul 13, 2003 3:52 am

Ideas u consider or learn when testing.

Post by Nathan » Tue Oct 26, 2004 3:27 pm

One thing I feel I have learned, and this is opinion, is that technical indicators add no value. Almost every phenominon defined by a technical indicator can be more easily defined by referencing by ohlc directly. channels, deviations, ratios.. Im not saying they can't work, but i for one do not think they help. for every system I tried to use an indicator on, i saw that the results improved considerably when I canned the indicator and based the system on a more simple relationship or measure.

Worse, by exessively abstracting from the data, it makes it difficult to get a feel for what is happening, or what the nature of the price movement is, so it is harder to build on a concept in a useful way. (at least for me).

I have recently been testing reversal ideas, in stocks, testing a basic entry with various profit target/stop points, and or time based exits. I was first testing this on Bond futures data for a bond system, but, tried it out on stocks too.

basic procedure being to:

1. test entry with time stops alone of various lengths
2. Test with equi distant profit - stop targets of various sizes, both
and symetrical and assemetrical.
3. Test combinations.

A random entry is used as a control.

I found that:

-testing on a basket of 100 stocks some amazingly simple reversal (long only) entries work quite well when combined with short holding periods.

-the small average profit per trade means that large gains are created by fast turnover. Smaller margins and high turnover mean the systems are more sensitive to transaction costs.

-many systems did theoretically much better than buy and hold over last 15 years, with skyrocketing returns since 2000.

- Based on the limmited work i have done, I would rather use a time stop than a price stop on a stock market reversal system .

-A small profit target combined with a short time stop created very good results.

-Like many good trend systems, a good reversal system is often elegant rather than complicated.

-I found that the logic of a a successful short term system is very different from a long term trend following system. Fast gratification seems to be better with a short term reversal system. (at least, based on the systems i created). Get your profit and bail, so u can free your capital up for other trades.

-It may be true that more technical sophistication is needed to overcome the issue of transaction costs, due to low profit per trade. However, buying into a decline does create possibility of positive slippage, so perhaps this mitigates part of the problem?

-After considering results of reversal systems and considering them in relation to trend systems, and how such traders are often on the opposite side of trades, I had a realization: they need eachother for liquidity. Both have a place in the markets, and i that is why both are viable concepts to work with.

If u have any free roaming ideas u have considered when testing, or results that were not what you expected, maybe u can tell us aobut htem here.

Bondtrader
Roundtable Knight
Roundtable Knight
Posts: 101
Joined: Mon Apr 28, 2003 11:07 am
Location: Zimbabwe

Post by Bondtrader » Wed Oct 27, 2004 11:29 am

I've learned that my psychological sweet spot contains longer term systems that only trade (futures contracts) once or twice a year. I've learned that enlarging my portfolio of traded markets is ALWAYS good, even when going from 30 markets to 60, and also from 60 to 90. I've learned that trading a big portfolio gives me the opportunity reduce the "dynamic range" of total heat, which makes CAGR more forecastable and which makes max-DD incredibly more forecastable.

Someone (Ted Anniman?) posted as much, a year or two ago, but it took me a lot of time and a lot of comparison testing to see just what he meant.

TC
Roundtable Knight
Roundtable Knight
Posts: 101
Joined: Thu Jul 08, 2004 2:12 pm
Location: Toronto, Canada

Post by TC » Wed Oct 27, 2004 4:37 pm

Bondtrader wrote:.........trading a big portfolio gives me the opportunity reduce the "dynamic range" of total heat.......
Bondtrader

Please expand on this concept, I'm not sure what you mean.

Also, if you are trading 90 futures markets I assume you included many non-US markets. If so, were there any special considerations you factored in to your portfolio selection eg exchange rate risk associated with non US-dollar denominated instruments ?

I have recently purchased 25 years of CSI's global futures data to both validate and reoptimise my current system as I am also inclined to expand the number of markets I trade before I diversify timeframe and/or strategies on my current US portfolio.

Thanks

Tom

Bondtrader
Roundtable Knight
Roundtable Knight
Posts: 101
Joined: Mon Apr 28, 2003 11:07 am
Location: Zimbabwe

Post by Bondtrader » Wed Oct 27, 2004 5:18 pm


Nathan
Roundtable Knight
Roundtable Knight
Posts: 127
Joined: Sun Jul 13, 2003 3:52 am

bondtrader

Post by Nathan » Mon Nov 01, 2004 10:17 pm

thank you for that interesting thought and for pointing out those past discussions. the concept of looking into the effect of average vs. maximum heat is very intesting.

on a more mundane note, One thought i had today was, "im glad I went back trade-by-trade to check the data" When checking out the resutls on one stock that had unusually positive results, enough to alter the entire system performance on a portfolio test. this being a buy on weakness and flip type of test, it had bought on a bad super-low tick then flipped on the real data going forward, creating one fictional monster profit.

In fact I think the recent ideas i have been looking into might be particularly vulnerable to hindsight bias, as they all involve buying on weakness and flipping out. Achieving some extraordinary backtesting results, yet, but Perhaps the stocks that would have killed the systems in real_time are bankrupt or something similar, so they don't show up in my
test.

Post Reply