I enjoyed the book, I found a lot of interesting things, expecially the idea of trading in terms of target returns and the best way to get it with optimal f.
I have the feeling that Vince dismisses Kelly too quickly:
In trading we can count on our wins being for various amounts and our losses being for various amounts. Therefore, the Kelly formula cannot give us the correct optimal f
pag. 122
I did my Kelly formula homework on Excel: the only thing I kept constant were returns on wins and losses, i.e. wins and losses don't have to be of constant amount when you risk a fraction of capital.
Kianti, Ralph is saying the Kelly formula is no help when there are a large number of different trade outcomes. Not (a large number of trades); rather, (a large number of trade outcomes).
Let me suggest that you apply your spreadsheet to the set of trade outcomes previously analyzed by Ted Annemann, here on the Trader's Roundtable forum, five years ago (link). There are 500 trades, and their outcomes (net profit or loss in R-Multiples) are:
5 trade outcomes (1% of 500 trades) are -3.0R
20 trade outcomes (4% of 500 trades) are -2.0R
25 trade outcomes (5%) are -1.5R
150 trade outcomes (30%) are -1.0R
50 trade outcomes (10%) are -0.75R
75 trade outcomes (15%) are -0.5R
50 of trade outcomes (10%) are +1.35R
50 trade outcomes (10%) are +2.0R
50 trade outcomes (10%) are +2.5R
20 trade outcomes (4%) are +5.0R
5 trade outcomes (1%) are +9.0R
As you can see from Ted's plots, the optimal-f is around 0.08 (betting ~~ 8% of equity on each play). Does your spreadsheet calculate an optimal f of approximately 8% ?
Per my reading of Vince, I believe that Opt F is defined and calculated as in the quote above but the same resulting number is also then used in trading as the fraction of equity to bet. Any comments from those who have used it?
Also is the thought that T-Blox should spit out Opt F as the fixed fraction that corresponds to the highest CAGR in a stepped test of % of equity. This never corresponds to my calculation of F (using Vince's Holding Period Return method.)
If I am mistaken in my understanding of Opt F, the question becomes, "How to convert Opt F to the fixed fraction of equity to risk?" For that is the number we all trade with.
I just use Parameter Stepping in Blox to try lots of different betsizes, then I look at the table of results and pick the one that "feels" like the best compromise between gain and pain --- to me.
Another thing you could do, though, is to step the betsize parameter, and then plot your results. You can find optimal-f on the plots; it's the betsize where CAGR% peaks, but (as some people have noticed), the drawdowns are punishing! Here is an example of this kind of work: viewtopic.php?t=5698 . Optimal f is marked with a blue dot. User "corvus" is perplexed that peak-MAR (green dot) occurs at a lower betsize than optimal-f.
Yes, Sluggo I agree that stepped parameter runs are the quick way to find the optimal risk fraction. What I want to know is why this number doesn't correspond to an F derived mathematically from the same trade log per Vince's procedure. My guess is that I'm missing something in my understanding of Opt F.
Incidentally, one of the salient points that Vince makes is the extreme drawdowns when anywhere near F. Great chart, by the way on your linked post. As you said, "a picture is worth..."
P.S. If anybody has Ralph Vince's email I'd love to pose the question directly to him.
sluggo wrote:Another thing you could do, though, is to step the betsize parameter, and then plot your results. You can find optimal-f on the plots; it's the betsize where CAGR% peaks ...
That's what I did in the f.jpg. Do you think it's worth the hassle?
TrendFriendly wrote:Kianti,
Referring to your f.jpg, What is the relationship between your Opt F of 23% and your "Bet Ratio" of 7.67%
Once the highest f is found, it can readily be turned into a dollar
amount by dividing the biggest loss by the negative optimal f. For example,
if our biggest loss is $100 and our optimal f is .25, then −$100/−.25 = $400.
In other words, we should bet one unit for every $400 we have in our stake.
In the f.jpg, Bet Ratio = 1 / starting value = 1 / 13.04