Making money with a losing strategy

[Mike805]

Quote from austinp:

Secondly, a losing strat is a net-loss at year's end. Whether you blend that or run it alone, the same result in part or whole is negative outcome. So there is no way to blend a losing strat with winning strat and experience greater degree of net profit than the winning strat alone.

Simply not true...

Let's assume you base your position size on the max drawdown of a strategy. Say for a 100k account you trade 1 unit and given 1 unit day-in day-out a 10% DD is bound to happen and a 10% return is likely.

Now say you're willing to tolerate a 50% DD to gain a 50% return. So you trade 5 units. Simple.

Now, suppose you design a strategy that hedges/gains on the days you have the potential for a large drawdown in the core strategy, but, this hedging strategy a loser overall. Now, your drawdowns are reduced to 5% and your return for the year goes to 7%. The cost of the hedging is 3%.

Again, you're still comfortable with a 50% DD. Assuming we can scale DD and gain linearly (big assumption and the real world doesn't work this way, but one can come close), now in theory you can trade 10 units and achieve a 70% return with a 50% DD.

This is a simple concept that quite a few funds employ... I am surprised that with your "years" of investigating systemic approaches you have never come across this scenario.

 

[goodgoing]

Quote from Mike805:

Again, you're still comfortable with a 50% DD. Assuming we can scale DD and gain linearly (big assumption and the real world doesn't work this way, but one can come close), now in theory you can trade 10 units and achieve a 70% return with a 50% DD.

Double leverage for a 40% increase in return is a recipe for disaster imo. However, it is a valid method I agree.

Quote from Mike805:

This is a simple concept that quite a few funds employ... I am surprised that with your "years" of investigating systemic approaches you have never come across this scenario.

He does retail hype...

 

[goodgoing]

Quote from jcl:

It is indeed so. Ralph Vince has analyzed the correlations ...

His analysis works only for constant dollar stop loss. In addition, future volatility and drawdown are big unknowns. His method is a recipe for disaster besides being a waste of time and too complicated for anyone who understands what is going on. Trading means moving big size at the least possible risk. Can you move 1000 ES per trade for 0.01% bankroll risk? Then we are talking real trading. The rest is for books and retail hype.

 

[Realistic]

Easiest way to make money with a losing strategy is sell a newsletter and cherry pick calls...

But of course that secret is out... considering all the newsletters already in publication....

 

[jcl]

Quote from goodgoing:

His analysis works only for constant dollar stop loss. In addition, future volatility and drawdown are big unknowns. His method is a recipe for disaster besides being a waste of time and too complicated for anyone who understands what is going on.

Hmm, until now I thought things are normally too complicated for people that do not understand what's going on, not the other way around. Anyway, do you want to say that portfolio management is generally "a recipe for disaster besides being a waste of time", or do you mean only Vince's algorithm? In that case, can you let us know which portfolio algo is not a recipe for disaster?

 

[amazingIndustry]

that is the biggest bogus I have ever heard. If strategy B has a negative expectancy, strategy B in combination with any other strategy(ies) can NEVER outperform a basket that does not contain strategy B. Now, I am not saying that strategy B may never add benefits in terms of limiting drawdowns or risk adjusted returns, but in terms of pure returns your claim is utterly wrong, lacks any mathematical or statistical foundation.

Edit: No correlations will change the total return claim I made, but correlations play a part in whether strategy B can reduce risk or not.

Quote from jcl:

Theoretically, it should be possible to make money even with a strategy that is not profitable. I don't mean selling it to newbies for $10000, but by using it in a compound system.

Suppose you have two uncorrelated strategies A and B, with A returning 80% profit and B 40% profit. A compound system of both strategies will not return 60%, but likely more than 100% - the whole is greater than the sum of its parts. I think this is commonly known. But surprisingly, this should even work when strategy B is slightly losing, f.i. -10%. As long as it has some negative correlation to the other strategies, adding it to a compound system can theoretically improve the overall return by reducing drawdown.

Has someone already made experiences with compound systems from uncorrelated or negatively correlated strategies and assets? What's the best money management for such systems - covariance based or optimal f?

 

[amazingIndustry]

it does not pull strategy B into any winning territory, the best that could happen is that drawdowns from strategy A in aggregate (as measured as a total basket of strategy A plus B) may be reduced because of temporary gains in strategy B. A strategy with negative expectancy can never return you more than a basket of strategies that does does not include such negative expectancy strategy in pure dollar terms.

Edit: I made no assumptions about correlations here, hence the term "may be reduced". But correlations do not change anything what I said total returns.

Quote from NetTecture:


He is right, possibly - if the signals from winning strategy a happen mostly at the time strategy b looses a lot, it could be used as a filter, which may pull strategy b into winning territory. Something along those lines.

 

[ssrrkk]

Quote from ssrrkk:

Mean and variances add for uncorrelated RVs. I don't see how -10% added to +40% will improve the results if they are completely uncorrelated: It should give you 30% return. Your variances will add too so it shouldn't help your draw down either. If there is a anti-correlated component, then the variance of the sum will be reduced by the covariance of the two.

Back to this: if you think a little bit further, this whole thread is completely meaningless. So if the two strategies are uncorrelated, then there is no benefit in mixing an inferior strategy. But if the two are anti-correlated, then the OP suggests it gives smaller drawdowns. But the problem is this: if there are any significant correlations or anti-correlations, then the two strategies are learning to recognize the same events. otherwise, the correlations will not be significant. So if you are basing your strategies on the same statistically rare events, then you are not learning anything new. There is no new information you are exploiting. Therefore, if the two strategies are correlated or anti-correlated, then it doesn't make sense to add them. Might as well just increase your weighting on the first strategy and you will get the same effect, minus the noise.

 

[amazingIndustry]

That is not entirely true. Yes the peak highs will be lower but the peak troughs may not necessarily be lower they could actually be higher depending on the correlation between the returns of both strategies. Fact, however, is that the total return when including a strategy with negative expectancy is gonna be lower, anyone who disagrees with I plead with to continue trading because anyone in the market less intelligent than me is welcome to stay in the game and periodically re-deposit new funds into their trading accounts.

Quote from austinp:

You happen to be mistaken on two counts

First, the days of selling anything for $10,000 related to trading ended years ago. There aren't any cash-rich newbies willing to toss out $1,000s every which way on a whim. This is 2012, and it's a whole new world in the trading realm.

Second, a losing strat is a losing strat. It cannot smooth your equity curve in a positive way. The peak highs will be lower and the peak troughs deeper. Who cares about what happens in between?

That's not much different than common assumption to reverse-engineer a losing strat to make it a winning strat by fading. Logical deduction, incorrect assumption in reality

 

[amazingIndustry]

lol, dont make me laugh out loud. You cant have copied this from someone else because its full of mistakes!!!

For beginners, if returns of strategy B were indeed correlated with a coefficient of -1 with returns of strategy A then you cannot possibly have a strategy B where the average win is 90, and average loss is 110 given your stated avg win and loss of strategy A

Quote from jcl:

I give you a very simple example.

Assume strategy A has 50% win rate, average win $200, average loss $100. Strategy B has also 50% win rate, average win $90, average loss $110. B is a losing strategy. Assume that both strategies do 100 trades per year and are 100% anticorrelated, i.e. the correlation coefficient is -1.

Now a little statistics for calculating the drawdown and the profit. We assume the trades are normal distributed and the probability of N losing trades in a row is thus:

p(N) = (50%)^N.

Let's assume a worst drawdown N at about three times the standard deviation of the Normal curve. 3*sigma corresponds to a probability of 0.0027. Thus,

(50%)^N = 0.0027
=> N = log(0.0027) / log(50%) = ~8.5

Thus, strategy A has an average worst drawdown of

8.5*$100 = $8500

and thus an annual profit of

(50*$200 - 50*$100) / $8500 = 58%.

For simplicity's sake I've omitted margin and commission here, so the $8500 are the required capital for trading the strategy. Now the same for the compound strategy of both A and B. As they are 100% anticorrelated, B always wins when A loses. The average worst drawdown is now

8.5 * ($100 - $90) = $850

and the annual profit of the compound strategy

(50*$200 + 50*$90 - 50*$100 - 50*$110) / $850 = ~470% !!

Adding a losing strategy increases the annual profit from 58% to 470%.

This is an extreme example, as normal strategies are not 100% anticorrelated, but you see that even a slight anticorrelation of a losing strategy can remarkably increase the profit of a portfolio. No higher math and no paradox involved.