No, it is the profit factor at the end of the testing/trading period. Expectacy is a random variable and a function of time. Profit factor measures how you performed at the end of the day. A key metric cannot be a function of time. I don't expect Van Tharp to understand this.
expectancy and profit factor are simply two different methods of expressing an edge (both can be expressed with the same set of variables). One is a difference equation, the other a ratio. As a previous poster mentioned, pf > 1 implies positive expectation. Likewise, for pf > 1, then E(rtn/trade) must be > 0. You can even relate the two if you want. E(rtn/trade) = Pw*Wavg*(1-1/pf) where pf > 0 Neither is more or less random than the other in the sense that neither predicts the future with certainty. The kelly paper is not very accurate for continuous outcomes. You can not simply use average rtns in the binary kelly optimization, the optimized fraction will be erroneous. It requires probability density functions , and related payout distributions (see some of R. Vince's work).
If I understand your point correctly, that you're talking about varying your position size based on traits within your p/l distribution, then yes I have done this before. Basically what I did & intend to do further is to say 'whats my expectancy for all trades with (or against) the trend? or 'whats my expectancy for all trades in a high (or low) volatility environment?'. I came up with what I thought may be common traits within the distribution and then tested the ideas. Also, to do this requires a large number of trades for each bucket you're trying to test. Once this is done, if some of the groups have a higher expected value than others, I would use a larger position size for these groups. This may also tell you that certain groups have a negative expectancy & I would deal with that as well. Expectancy by itself is useful, but there is a lot more to be learned when you dive into it more. Its not dissimilar to reading a statistic that says '50% of marriages end in divorce' & then once you peel the covers off you realize that a much higher percent of 18 yr olds getting married get divorced than 28 yr olds. Eric
The math on both of these measurements is pretty much the same. Both numbers describe a data set. They are dynamic & change when you change the set (i.e. add more trades). I think when used correctly either one would tell you useful information for decision making. Eric
Not entirely correct. Both numbers describe a boundary condition and can be the outcome of an infinite number of data sets. This is just one problem. Ron