Future does not confirm the past!!

[nitro]

Quote from gifropan:

Why is is that most of the time when looking at charts and backtesting a strategy it seems to work, however, when applying it in real time for some reason it breaks down. This is a real puzzle. Anyone else experience this?

The key to trading isn't finding things that work, it is finding a model that adapts to market conditions, for reasons you point out. When you find something that works, you have discovered a slice of truth, assuming causation. Experienced traders know how to shift regimes within a model, or shift to a totally different model, as markets evolve.

1) If you want to have a fully automated model, you have to have some sort of way of figuring out adaptation and regime shifts. Google both of those. Regime shift is often called a phase transition.

2) By far, and I mean far, the most popular way that dealing with shifting dynamics isn't to try to fix the model, it is to give it to a trader that understands how the model evolves under different market conditions, and then he trades around that. So the model becomes a crutch to a human trader, instead of a dictator whose rules are followed to the letter.

Most would far prefer option one above, but it is probably close to impossible to program.

 

[ScoobyStoo]

Quote from nitro:

The key to trading isn't finding things that work, it is finding a model that adapts to market conditions, for reasons you point out. When you find something that works, you have discovered a slice of truth, assuming causation. Experienced traders know how to shift regimes within a model, or shift to a totally different model, as markets evolve.

1) If you want to have a fully automated model, you have to have some sort of way of figuring out adaptation and regime shifts. Google both of those. Regime shift is often called a phase transition.

2) By far, and I mean far, the most popular way that dealing with shifting dynamics isn't to try to fix the model, it is to give it to a trader that understands how the model evolves under different market conditions, and then he trades around that. So the model becomes a crutch to a human trader, instead of a dictator whose rules are followed to the letter.

Most would far prefer option one above, but it is probably close to impossible to program.

Wise words Nitro.

An automated strategy that is not totally adaptive is doomed to failure in the long run. However, if you can build an automated strategy that defines all its parameters based on actual market data, rather than some random numbers spat out by a backtest optimisation, then you are heading down the right track.

And you are right, it is complex and very hard to program (hence why hedges funds pay massive $ to grab the best software developers).

Also, it never fails to amuse me that so many people on ET assume that just because nobody on these boards is telling everyone what they've done and how they've done it, that it is impossible.

Naive in the highest order...but then again, that's probably why they are known as 'sucker money'.

 

[MohdSalleh]

trading is chasing your own tail round and round in a circle

 

[travis]

Quote from IShopAtPublix:

Financial markets will never be "solved" by quantifiable measures. As you can imagine, all kinds of sophisticated statistical analysis on super computers has not resulted in a "break through"

That's like saying that markets are not predictable (random walk hypothesis). Then why are you trading at all? Or do you think that your mind isn't (even if unconsciously) doing what trading systems do?

 

[ProfLogic]

Quote from gifropan:

Why is is that most of the time when looking at charts and backtesting a strategy it seems to work, however, when applying it in real time for some reason it breaks down. This is a real puzzle. Anyone else experience this?

As long as your charts are based in increments of time, range or transactions, your results will be inconsistent.

 

[ivanbaj]

Quote from travis:

That's like saying that markets are not predictable

Yes! Past does not predict future. Now it is all there is.

 

[travis]

Quote from plyka:

I think you're asking why you can back test a strategy and it works great but then when you actually trade said strategy it breaks down, correct?

Well i'm a discretionary trader so this is out of my area of expertise, but from what i've read it has to do with what exckhart calls degrees of freedom. When you come up with a strategy (mechanical one) and then backtest this strategy, the more you add in parameters into the strategy the more you are optimizing it for the past.

For instance, say you have a very simple strategy --20-day breakout with a stop at the low of the breakout day and and profit target of 20% rise. This has very minimal parameters. If you backtest this, it should more or less equal future results. However when you start optimizing this strategy with more and more parameters then all you are doing is creating a system that fits well with the past.

For instance say that you test it out and it is more profitable if you put in the parameter that you only take trades on Monday/Wednesday/Friday. Well it could be that in the past it just coincidentally happened that the breakouts which happened on M/W/F worked better. That is, there is no viable reason it worked better, just happenstance.

Well if you add in this parameter then the backtesting is going to look much better than the real time trading. Because you are optimizing the system. You tested for Tuesday and Thursday and it just happened that breakouts during those two days failed more in the past.

The more parameters you put into the system, the more you are optimizing the system based on what actually happened in the past. You are, whether on purpose or not, taking only the good days of the past and eliminating the bad days.

If there is a legitimate reason why breakouts work better on M/W/F then there is no problem. HOwever if it was just coincidence then this parameter is going to cause problems.

This is very very good and clear. Especially considering it's coming from a "discretionary trader".

 

[travis]

Quote from ivanbaj:

Yes! Past does not predict future. Now it is all there is.

Then how are you, too, hoping to make money in the markets? Randomly picking your trades? Your trading experience doesn't count, if "past does not predict future". Unless you meant "past does not ALWAYS predict future", in which case we all agree, but then it's also granted.

 

[ivanbaj]

Quote from travis:

Then how are you, too, hoping to make money in the markets? Randomly picking your trades? Your trading experience doesn't count, if "past does not predict future". Unless you meant "past does not ALWAYS predict future", in which case we all agree, but then it's also granted.

I can predict the future (short term) by using now. Our past P&L does not count. There is hope for all.