Systematic Traders - How many systems do you run?

[achilles28]

Quote from jcl:

That other poster was probably me, but you're right with the preference of uncorrelated markets. Problem is, when you trade 20 markets, each of them is most likely correlated to 10 of the others. My point was that you can trade them nevertheless. In the worst case you'll get the same return: trading two 100% correlated markets will just produce the same profit when you assign 50% of your capital to each of them. If the markets are less than 100% correlated, which is normally the case, you'll improve your return.

I think we're both saying the same thing, but from different angles. Splitting capital between 2 correlated markets results in equal (or better) performance had the capital been allocated to one instrument alone. However, two correlated instruments relative to 2 uncorrelated instruments, the obvious choice is two uncorrelated. The trick is the assumption of 1% risk per trade, or whatever, across 10 instruments, each of which might be highly correlated, producing a bunch of signals at once = far more than 1% risk per trade.

Quote from jcl:

Normally you would not rank strategies by performance, but calculate capital assignment factors from the equity curves of the separate strategies. There is no closed formula for those factors. You need a computer optimization, but it's fairly simple. Ralph Vince described the method in his books.

Not sure what you mean by capital assignment factors. Can you explain? Why not rank by performance? I would. Seems most logical.

 

[jcl]

Quote from achilles28:

The trick is the assumption of 1% risk per trade, or whatever, across 10 instruments, each of which might be highly correlated, producing a bunch of signals at once = far more than 1% risk per trade.

With 10 totally correlated instruments the risk would be 10 times higher, but also the profit, resulting in the same return (which is defined as profit divided by bound capital).

Quote from achilles28:

Not sure what you mean by capital assignment factors. Can you explain? Why not rank by performance? I would. Seems most logical.

Because a rank alone tells you nothing, you need a quantitative factor, an optimal percentage of the available capital that you can assign to that strategy. This percentage also does not depend on performance alone, but also on correlation, drawdown, and maximal loss, so you need the whole equity curve to calculate it.

I can't explain this here in short, but you can find more on the web when you google for "OptimalF", "Kelly" and "money management".

Quote from HurricaneUS:

it's widely known that intradaybill is another wannabe perennial loser who's only here to hawk that curve-fitting software priceactionlab that 's only good for making inferences from small sample sizes (like 15 results...i'm not kidding)

just check his post history..almost every other post will have some reference to that piece of shit software he so ardently believes in...

In defense of intradaybill, I think that he's real, not like all those people that register from time to time, make some posts about how great "priceactionlab" is, then disappear again. The guy selling that software goes apparently to great lengths to promote it. But I suspect that intradaybill is a real customer who has really paid $10K for that software, believing that it would make him rich. This might explain his bizarre posts.

 

[logic_man]

Quote from masterm1ne:

My man... I've found through nightly study of markets, one could use 1 pattern and make a very good living in most any markets. The only problem is it requires specific conditions.

Why do you find the requirement of specific conditions a problem? Does it lead to too few trades? Even across a wide number of markets?

I'm working across multiple markets with the same strategy as well and I'm finding that at any given moment, at least one market has triggered a trade, but there are times when I have no position in any market. I suspect that if I continue to add markets, my percentage of time in the market across them all will get closer to 100%.

 

[jcl]

Quote from intradaybill:

Most including you here are some type of a programmers with limited computer knowledge who hope to one day find a cheap way to get rich through some programming miracle. While you and your knind fart here every day multiple times and you have made the atmosphere in ET unbearable, I make money. My partners and I have spent over $100K for software over a 10 years period. You and your kind are miserable wannabes and you will never make it trading because you do not have what it takes for that:

- Money
- Team
- Tools
- Environment

Instead this is what you got:

- no money
- no team
- no tools, just playing with programming language
- you operate from your bedroom or garage

Well, in the anonymity of a forum you can of course make any claims about yourself. We can only judge you from your posts here. They never contained anything substantial - only either platitudes, or insults, or advertisements for that scam software. Maybe you're a great super trader, but you're hiding this very successfully.

As to me, anyone can just google me under my real name, Johann Christian Lotter. On the web you can find easily my address, my companies, our team, the projects, even the trading tool and equity curves. If you want to appear here as the super trader, then show something substantial - or just shut up and save bandwidth.

 

[braincell]

I have about 20 excellent systems for each of the scanned 930 stocks, approximately 19000. They are autogenerated, and in total, my CPUs have produced well close to a billion strategies by now. I run only 14 (up to 20) on real money because of the correlations (ie diversification) and capital, but it's enough to make my portfolio equity curve look as smooth as a baby's bottom. Now if only I could get the SEC to allow 10:1 margin for overnight equities... :/

 

[gmst]

Quote from braincell:

I have about 20 excellent systems for each of the scanned 930 stocks, approximately 19000. They are autogenerated, and in total, my CPUs have produced well close to a billion strategies by now. I run only 14 (up to 20) on real money because of the correlations (ie diversification) and capital, but it's enough to make my portfolio equity curve look as smooth as a baby's bottom. Now if only I could get the SEC to allow 10:1 margin for overnight equities... :/

LOL on last 2 lines. Congratulations on getting such good results. Do you mind posting few (maybe 2-3) of your single system equity curve and the portfolio curve? That would be instructive!

This is a request to all the systematic traders who are participating. Please share your portfolio construction methodology along with portfolio equity curve and couple of single system equity curves for comparison purposes. If you have not managed to develop a portfolio construction methodology yet, this exercise would be very useful in developing a methodology that is required for long term success.

 

[gmst]

Quote from achilles28:

My only experience with this problem is theoretical. The way I look at it - prioritize capital allocation according to performance. Rank the strategies according to performance (monthly, semi-annual, annual, 5 year....whatever) then give trading priority to higher performing strategies over lower, if you run into capital constraints from multiple signals. For instance, if you've got enough capital for 3 trades, and 7 signals were produced by strategies ranked #3, #49, #19, #18, #8, #23, #12 (according to their historical profitability), take signals from #3, #8, and #12. This maximizes capital accumulation and minimizes draw-down (theoretically). The second filter to all that is correlation. My take on correlation (instrument and strategy), is that uncorrelated strategies and pairs should take priority over correlated instruments and strategies. If you trade the ES, YM (which are highly correlated), and CL (which is less correlated), and three entries are triggered, it's better to take the ES and CL, then the ES and YM. The reason being, losses get compounded in correlated markets. Not sure why another poster said they don't. If you think about one strategy triggering a buy for GOOG, ES, APPL and ER2, and the market tanks, chances are, all four positions will take a hit. Stocks and the indices are tightly correlated. But if you were triggered long for the ES, 10 year, Bund, and the Euro and the market tanked, chances are, one or two of those trades would work out, as some of those instruments exhibit weak correlation. This minimizes your losses and protects your equity curve, which is number#1.

As far as nitty gritty stuff, trade size should be largest for your best strategies, and titrated down as strategies become less profitable. You want to maximize capital accumulation so weigh to your big guns first, 2nd best 2nd, 3rd best 3rd etc. Ranking can be done on a risk-adjusted basis with 1 contract, over a predefined look back. Or an average look-back. No need to curve-fit. It should be obvious, which are the best, which are second tier, third tier etc. I find it helpful to think in terms of Tiers.

I've got around 15 strategies that work and plan to add 2 or 3 variations to each "base" strategy. Pyramiding can be very profitable. Every system, if you think about it, is trend following, because it picks direction (long or short). If the strategy is profitable, why not add to it? So I've got another stack of ideas for pullback entries. These are abundant and not really something that requires a lot of thought. U can add on pullbacks, on candle breaks, volume breaks, formations, trendline breaks, oscillator crosses etc. Doesn't matter. If you strategies are solid, and pick direction well, you should be able to find logical, easy, plentiful places to ADD to that position. This compounds winners. Just some thoughts.

Thank you very much! Great Response! I have started working on developing a position sizing and portfolio construction methodology, that has kept me very busy so responding late. I am approaching portfolio construction of systems in the following way:

1. Ranking each system individually either using a combined rank using parameters like PF, Sharpe, Win/Loss, DD, Smoothness of curve etc. Or using multiple dimensions to construct a multi-dimensional rank for each system. An E.g. for a single rank would be:

http://unicorn.us.com/trading/expectancy.html

2. Coming up with a position sizing methodology for each system viz. fixed fractional, fixed ratio, volatility based etc. I would also do MonteCarlo in steps 1 and 2 to generate multiple instances of trade sequences. A bit of work but important and required for robustness.

3. Compare correlations of systems on a historical basis (using daily/monthly data) and then on an ongoing rolling basis.

4. Deciding on position sizing different systems to construct a portfolio. Using an optimization procedure sounds right, but imo, is wrong. Instead, position sizing of different systems should be based on Monte Carlo again.

After the portfolio has been constructed - On an on-going basis, following measurements would be required:

1. Checking if edge persists for each system
2. Correlations between systems
3. Changing Position sizing for different systems in the systems portfolio periodically based on above two points.

 

[masterm1ne]

Quote from logic_man:

Why do you find the requirement of specific conditions a problem? Does it lead to too few trades? Even across a wide number of markets?

I'm working across multiple markets with the same strategy as well and I'm finding that at any given moment, at least one market has triggered a trade, but there are times when I have no position in any market. I suspect that if I continue to add markets, my percentage of time in the market across them all will get closer to 100%.

Trade # was not the problem I was referring to LM. The problem I was referring to is a dynamic judgment ability I think being a successful trader requires. This means you have to monitor markets you are trading or have some sort of basic alert for the pattern, and then you have to check it youself if you want to place a trade. This can get tedious the more markets you watch.

 

[gmst]

Continued from above.....especially the point 4 above - which is super complex imo.

In my view, the biggest challenge is - how to model combined DD for the portfolio. This is the real risk on portfolio level going forward. This is complicated because say you have 10 systems and you run 100 simulations (bare minimum imo) on each system to generate DD figures for each particular system. To know the portfolio DD from these individual Monte Carlos, you will need to do 100^10 = 100,00,00,00,00,00,00,00,00,00 calculations.

Doing even 10 simulations per system, would mean 10^10 = 10 billion calculations to compute portfolio DD. However, this simple approach (even though computationally time consuming) doesn't take into account correlations between the systems and hence is wrong!

The correct way to approach this problem, again imo, is to run a Monte Carlo on all the systems together, where you are sampling from a joint distribution using a correlation table. This will allow us to compute usable statistics for the portfolio DD etc. using 'just' 1000 simulations. Makes it computationally much easier to solve. The drawback in this case is that this assumes we know the 'true' correlations between systems and the correlations between systems are constant. We know in reality both these assumptions are wrong.

Thinking further - It seems like the "correct" way to backtest a portfolio of systems over multiple years would be to compute rolling correlations every year or every half year and use this rolling correlation table to run Monte Carlo on different systems using a joint distribution. Portfolio of systems will be re-balanced in this method every year or every half year.

To be continued....

 

[Rationalize]

Quote from braincell:

I have about 20 excellent systems for each of the scanned 930 stocks, approximately 19000. They are autogenerated, and in total, my CPUs have produced well close to a billion strategies by now. I run only 14 (up to 20) on real money because of the correlations (ie diversification) and capital, but it's enough to make my portfolio equity curve look as smooth as a baby's bottom. Now if only I could get the SEC to allow 10:1 margin for overnight equities... :/

10:1 leverage on equities is available synthetically. Nothing remarkable there.