There's something to be learned from anyone with a lot of mkt experience. I can think of different ways to do a Riemann-like analysis (so that it wouldn't be "mumbo jumbo") and many benefits from doing it. The only question in my mind is how to get started and would the effort be worth it. But I believe it would as it would expose the boundaries of your system if mkt conditions ever change...It seems proactive instead of reactive to me.
How far back is enough for a backtest?
- Thread starter markc
- Start date
There's something to be learned from anyone with a lot of mkt experience. I can think of different ways to do a Riemann-like analysis (so that it wouldn't be "mumbo jumbo") and many benefits from doing it. The only question in my mind is how to get started and would the effort be worth it. But I believe it would as it would expose the boundaries of your system if mkt conditions ever change...It seems proactive instead of reactive to me.
No no,
You misunderstand me. While it is possible that advanced mathematics may one day light the corners of theoretical trading, what I meant to say is that I learn from the way that the _history_ of math has proceeded from an essentially ad-hoc methods made to serve science, into an a discipline that stands on it's own right that has proceeded from the singular examples to the abstract.
My point is that trading system developement has the feel of mathematics around the time of Euler. That is, it is just a bunch of indiscernable clever hacks without any unifying principles.
nitro
Well, of course I love that. Anyway you can weave the history of mathematics and science into trading is all right by me...
I don't think anyone can dispute that testing of an idea is worthwhile. Why trade an idea if its never worked in the past? Some of the successful hedge funds that are built on the idea of backtesting ideas include Rennaissance, DE Shaw, all the trend followers out there, and so on.
For me, I like to make sure that any system works and has many examples in the following types of periods:
a. up markets
b. down markets
c. rising interest rate periods
d. decreasing interest rate periods.
e. rising VIX periods, decreasing VIX periods.
This pretty much takes me to early 1999. I don't really believe in data before that because the markets, indices, and market players were very different. For one thing, trading during the day right now is much more dominated by hedge funds than any period before.
For me, I like to make sure that any system works and has many examples in the following types of periods:
a. up markets
b. down markets
c. rising interest rate periods
d. decreasing interest rate periods.
e. rising VIX periods, decreasing VIX periods.
This pretty much takes me to early 1999. I don't really believe in data before that because the markets, indices, and market players were very different. For one thing, trading during the day right now is much more dominated by hedge funds than any period before.
there is another problem. i trade the es. through the year 2000 i backtested and traded the big s&p. there was a definite change in pattern when they changed the big s&p from a $500.00 contract per point to $250.00. BIG PSYCHOLOGICAL change. then When every one started switching to the emini, backtesting the emini was impossible before the year 2000 because it was so volatile and non liquid. after 2000 the emini was ok to backtest and you would not get as many false signals. But backtesting is VERY necessary. If it won't work in the past, I would not bet a penny it would work in the future.
pat
pat
I backtest using Stock Watch Pro's http//www.crontech.com/swp backtester. Although the history repeats itself, you should always take any backtesting results with a grain of salt and apply a reasonable doubt when trading in real life. With that said, I think backtesting is a necessary step for any trading strategy.
AMEN. Find yourself some software or manually monitor forward testing results.
If you'd like validity to your results, I'd look at the number of trades. 10 trades in a 10 year period provide no validity. However if you are trading 100 times a day throughout that period, there may be some validity to your system. Regardless, I would give forward testing a whirl.
Here is the sort of example of the kind of abstraction I am talking about.
The "argument" may go something like this:
First assumptions: APT describes the market at the high frequency, that is, arbitrageurs instantly bring price aberrations back inline thru arbitrage.
Now "Theorem":
Given APT, no moving average crossover system that has as it's inputs lengths two twin primes (primes that differ by 2), where the time frame of the bars used to calculate the MAs are less than the product of the two primes - 1 (in ticks, minutes, hours, whatever,) can return more than 100% of capital invested
While this is pure fantasy, this is the kind of "abstraction" I am talking about.
nitro
The "argument" may go something like this:
First assumptions: APT describes the market at the high frequency, that is, arbitrageurs instantly bring price aberrations back inline thru arbitrage.
Now "Theorem":
Given APT, no moving average crossover system that has as it's inputs lengths two twin primes (primes that differ by 2), where the time frame of the bars used to calculate the MAs are less than the product of the two primes - 1 (in ticks, minutes, hours, whatever,) can return more than 100% of capital invested
While this is pure fantasy, this is the kind of "abstraction" I am talking about.
nitro