Expectancy above .50??

Gentlemen,

I highly appreciate all of your thoughtful replies.

I’m wondering about this, many are talking about the win %, and not enough about the expectancy. As I found out with most strategies that take outright positions, the Win:Loss ratio is Inversely related to the Win %, meaning when one tries to improve their Win% they are giving up some from the Win:Loss ratio, and vise versa. So I don’t see the importance of having something that is over 50% or higher, I wouldn’t mind trading something that is 30% profitable if the Win:Loss Ratio was 4:1 for that system. Here’s the graphical representation of this relationship:



So I don’t see the particular importance of having something that has a high Win %, unless its for the sake of handling the psychological implications of losing trades.

Hardyards,

A strategy with a 66% win rate and a 3:1 W:L ratio would yield an expectancy of 1.64. That is amazing if you asked me. Something that does 50% with 2:1 only has an expectancy of .50 and I would consider that a very tradable system because most pros are believed to have expectancies about this .50 area aren't they?. So I don’t understand, is there something I’m failing to see here? And I think I might understand what you mean by imposing the strategy on the market. Maybe I need more filters to take fewer trades on the wrong time and more trades on the right time?

Steveyd,

I think this is exactly what I was looking for, so you trade a system that yields an expectancy of less than .50. The highest I’ve ever seen on any of my strategies after slippage, commissions and out of sample data, is .39 and that’s only doing about 50-60 trades on average per year on a 63 month run. So the numbers I’m coming across might not be unusual for a profitable system. Steveyd, how close are your numbers to .50 if you don’t mind me asking?


Again, thank you everybody for your helpful responses,

shane

 

My two cents on this (and there is no "right answer" anyways).

This is where reality meets simulation: If you're running at 30%winners you're counting on few winners to be very large. This puts your simulation at risk, since those 30% tend to happen when you're sick, traveling or have computer problems

Speaking from experience: you're much more likely to have reality match simulation if %win > 50%.

As a test: try to remove 20% of your winners in your 30%win/ 4:1 system. Or randomly remove 20% of all trades. It will most likely not be pretty.

 

Shane,

Here are the expectancies for 3 intraday systems that I am currently trading:

Contract Expectancy Trades/Year %Win
ES 0.20 300 75
NQ 0.20 133 73
CD 0.49 120 59

These numbers are based on backtesting on the last 2 years of data. I add some real-time discretion to try and beat or at least match these historical results.

 

dare to contradict in several ways.

1. profit factor is a very flawed concept. it does not account for any smoothness of the equity curve. you could have a profit factor of 2.5 and still prefer a system that shows half of that. frequency of trading is the thing that is missing plus time, spells smothness calculated by standard deviation of returns.

2. hitratio is literally meaningless for a system trader. only the combination of hit ratio and payoff ratio is relevant and each of them has hardly any meaning on its own. of course, i am talking system trader here, thus no psychological component influencing the trading decision.

3. draw down without reference to return is not useful when the underlying is subject to leverage. if i have a system that yields 100% pa i might well bear a 20% draw down. and i could lever down to 10% easily, so the draw down alone does not tell whether a system is preferable to another.

peace

 

you're right; this can be looked at in many different ways.

I find profit factor plus %Win to be two good dimensions since they together produce the expectancy and you furthermore can see how good your edge is in two different ways.

Re drawdown: I always use full face value of e.g. ES, not margin, since that's what you're taking loss/gain on.

Personally, I size the positions to ensure that maxDD stays under 10% for any strategy; but that's just me.

 

Think of it this way:

If you implemented this system in real money, when would you cut the system?

Statistical criterias greatly depends on the type of system you develop.

 

Here is another example. Now as discussed before, the previously shown results of strategies do not present what one would call tradable results. Here is another strategy that, based on the numbers, I would consider very tradable. The only downside is the number of trades in comparison to the previously shown strategies. It appears the less trades a strategy takes, the better its numbers look. Although these results are obtained using out of sample data, and includes slippage, my main concern with this setup is the statistical significance of data. So far for the forward tests it has been performing well. But can one really know when the window of stationarity for the strategy will shift? Or that it has shifted? Specially when using strategies that are based off of lengths? Any thoughts on this?

Thanks,

shane

 

I'm not familiar with TS, but I know Wealth-Lab has a monte-carlo simulator you can run testing on... maybe TS has something similar?

 

nope ^_^

(unless you want to program it like I did )

 

I tend to agree that Profit factor is a very flawed concept.

My experience as a purely mechanical trader has told me that if all parameters in the system are optimized over a finite number of trades, as the system walks forward in time the PF will drop. It's very debatable, but one hypothesis might suggest that IF all parameters remain fixed that as the number of walk-forward trades tends to infinity the PF will tend towards at best 1.0.

My strats that trade on longer time frames will see less trades in a given amount of time, and will typically yield higher PF (2.0 and above) within a window of time in which the strat is optimized. However, these strats can have awful equity curves that whip up and down and produce horrible consistency of returns.

I have some strats that trade on 15 min. to 60 min. time intervals, and take 12 to 20 times the number of trades in the same time interval. The PF might be anywhere from 1.3 to 1.6, but the consistency of returns (I measure by MS ratio) is significantly better and the resulting equity curve is smooth and upward trending.

I'd be very careful in using PF to judge a system, simply because:

1) It tends to have a dependency on the number of trades taken in the evaluation period, and

2) It provides no measure of the quality of the returns over time.

RoughTrader