Expectancy above .50??

shanoballs · May 7, 2006

Hi, Iâve been working on developing mechanical strategies for about a year and a half now and it appears that expectancy numbers that are greater than .50 or Sharpe ratios that are greater than 1 are almost impossible to come by after including slippage and commissions to the trades. What are realistic numbers that one should look for in developing trading strategies? I mean if exp. numbers that are greater than .50 are unattainable, I donât want to waste my time looking for such systems. It seems like net profitability is possible, but with exp. numbers there are much lower than .50. Is having simply a positive expectancy the bottom line to net profitability? Are there any day traders out there that are capable of 1+ Sharpe ratios? Or is it simply expectancy that is greater than 0 which is contributing to the net profitability in actual trading?

Here are some of the numbers that I am able to achieve as far as in and out of sample historical data go, they are far from holy grails, but positive expectancy:

Thanks in advance,

Shane

knifecatcher · May 9, 2006

I haven't done this for a while. It seems like the systems can only keep about 10% from the average winning trade. I am not sure about the nature of the execution. Have you tried trading these systems? It could be a little tough in real life when you might get only half the performance (for example only trade thru orders happen in real life while Tradestation assume you got filled and other problems). My current feeling is if you spend time honing your discretionary skills, you can go much further than statistics from simple systems but that is a totally different path.

Good trading.

Steveyd · May 10, 2006

How are you calculating expectancy? To me it appears all three systems have a > 0.5% expectancy.

shanoballs · May 10, 2006

Quote from Steveyd:

How are you calculating expectancy? To me it appears all three systems have a > 0.5% expectancy.

I'm using [(1 + (avg.Win:avg.Loss))*(%profitable)]-[1]

--OR--

[Avg.Trade Net Profit ]/ [Avg.Losing Trade]

which is the same thing.

and yes these strategies are of positive expectancy, how is your expectency is followed by a % symbol?

-shane

vikana · May 10, 2006

In my experience there are only really a few things you need to watch out for:

1. Profit Factor. Should be > 1.5 for a tradable system
2. %Win. I like this > 50%
3. DrawDown. Try to keep drawdown less than 10% of max equity (with whatever position sizing you choose) at any time and CHECK the duration of the drawdown. If e.g. the duration is 1 1/2 years, can you stick with it?

finally, try to convince yourself that you don't have too many parameters in the system. I'd say 1 to 2 conditions per main dataseries should keep you safe.

radolym · May 10, 2006

Quote from vikana:

I'd say 1 to 2 conditions per main dataseries should keep you safe.

What do you mean by this? Can you give an example with 2 conditions vs 5 conditions?

vikana · May 10, 2006

Quote from randomly:

What do you mean by this? Can you give an example with 2 conditions vs 5 conditions?

Sure, i just mean that there only is so much information you can extract from one data series, such as the price for instance

Two conditions could be long Entry if

c > mov(200,c) AND
rsi crossing above 20

My point is that if you have entry conditions such as

c > mov(200) AND
rsi crossing above 20 AND
macd trending up AND
mov(50) > mov(200) AND
c > c(5)

you could easily be doing a "hidden optimization", by filtering out trades that happen to not satisfy all conditions.

In a highly simplified way of looking at things, think about it this way: Each dimension (in a mathematical sense) allows you one independent variable = condition. So unless you believe the (fractal) dimension of the price series is larger than 5 you should not have 5 conditions, or you'll most likely have overfitted.

Steveyd · May 10, 2006

Quote from shanoballs:

I'm using [(1 + (avg.Win:avg.Loss))*(%profitable)]-[1]

--OR--

[Avg.Trade Net Profit ]/ [Avg.Losing Trade]

which is the same thing.

and yes these strategies are of positive expectancy, how is your expectency is followed by a % symbol?

-shane

I was calculating expectancy as a percentage of account size required which I find more useful than per dollar risked.

For daytrading, the expectancy is useful only in the context of frequency of trades, imo. ie. does the expectancy * frequency of trades make the system worth trading.

I am currently trading a number of systems with expectancy per dollar risked (as per your/Tharp's definition) less than 0.5, that are profitable and worthwhile due their frequency of approximately 200 trades per year.

knifecatcher · May 10, 2006

Quote from vikana:

In my experience there are only really a few things you need to watch out for:

1. Profit Factor. Should be > 1.5 for a tradable system
2. %Win. I like this > 50%
3. DrawDown. Try to keep drawdown less than 10% of max equity (with whatever position sizing you choose) at any time and CHECK the duration of the drawdown. If e.g. the duration is 1 1/2 years, can you stick with it?

finally, try to convince yourself that you don't have too many parameters in the system. I'd say 1 to 2 conditions per main dataseries should keep you safe.

Wow. You made it all clear with few sentences.

hardyards · May 10, 2006

Shanoballs ( interesting title!)

I gave away this type of mechanical approach years ago because trading the markets consistently requires a high degree of "rat cunning" and I could never programme it.
However, some good news first.
With a %A of 65% on both the long and short side, you are on a good path.
Now you need to concentrate on your Win/loss ratio, which quite frankly stinks.

If your %A >66% and your Win/loss ratio > 2.5 ( 3.0 is better), then you have yourself a workable system.
Take a look at the number of consec losing trades also ... looks to me as though you are trying to impose your system upon the market.