When developing an algorithmic trading system I always start by optimizing trade statistics (constant $ position for stocks, single contract for futures) and then layer MM to make the system tradeable with acceptable financial statistics (drawdown, avg trade vs slip/commisions etc.).

If the trade statistics are constant (they're not) Kelly will optimize geometric return at the expense of horrific drawdowns. I use MM to dial-in acceptable drawdowns. MM can't change the trade statistics, only lever (compound) or delever (reduce exposure) them into the equity curve.

As the number of %winners goes down the probability of consecutive loosers rises exponentially. This is why pure trend following CTA's are very computer execution driven (40-50% Winners, 1.2-1.4 Avg $Win/$Loss) layer timeframes and diversify lots of contracts. It's very hard for discretionary traders to stick to these systems during drawdowns, and without many trades/year (to recover) a DD can last for years independent of MM. Another example is levered funds selling premium in the options market with 80% winners. I've seen these funds go for 7 years, then boom! go bankrupt in a single 1987, 2001, 2008 market event. Buffett's "Pennys in front of steamrollers."

In engineering parlance the "trading system" and "bet sizing system" are two different transfer functions that are convolved when put together making things more sensitive to things staying statistically stationary (which they never are).

Ralph Vince has advocated switching from constant position to fraction of account after a certain number of trades. The number of trades he calls "Threshold to the Geometric". His work was to generalize Kelly for non-identical win/loss betting. It is still optimizing geometric growth with huge drawdowns. His math is right but my experience is that the statistics of real systems are alot more variable than people admit (examples above).

In summary, design the trading vehicle, entry-exit, stops etc for best (repeatable) risk-reward, and design the MM to keep the drawdown and recovery times acceptable. Then see what you're equity curve growth rate is. Sorry if it's not get rich overnite ;-(

Great points @Jeff!! thanks for sharing!!

In my opinion that comes from experience virtually any money management method that recalculates a trading size for every trade eventually leads to psychologically unbearable drawdowns. I can recall at least three times when clients stopped trading because of a drawdown level and duration that was unacceptable for them — although before that the same money management generated incredible returns like over 100% in 3 months. So it's mostly a matter of personal preference and true personal psychology; if you are able to withstand a drawdown of 60% that lasts for 8 months then you may eventually be rewarded with an incredible return — or, of course, a completely ruined account.

Therefore I believe that for most of people another approach is far more acceptable, although it works against common psychology: increase trading size in drawdowns, given you're able to identify those that are suitable for this purpose. This would recalculate the position size only at selected points, not for every trade, and this is the only method that may help reducing drawdowns and (what is even more important) reduce the recovery time.

All in all, don't do what most people do, and don't let your clients do it: to increase exposure at new highs of the equity curve and reduce exposure in drawdowns.

Personally, I think people are too hung up on draw down. The problem with draw down is that it is partially the result of random distribution. Now I have to admit that it’s less of a problem for me because I don’t trade manually. Yes, it still hurts a bit to see a 35% draw down but if the other stats are looking good then I leave it alone. And yes, it’s true that trading has been a bit of a rollercoaster ride for me the past two years.

I’ve tested a version of the method that Alex described on my own strategy, and it wasn’t satisfactory for me. Yes, it did lower the RDD but it also lowered the profit/RDD. So in the end it would be better just to risk a smaller percentage of the account per trade, which would lower the RDD but keep the profit/RDD intact. But that was on my strategy that trades only one instrument and has only one open position at any time. The results might be very different for a strategy that holds multiple stocks at a time.

if you have a system that is in need of stops then i suggest that you may not even really have a system that is going to endure transitions from trending, reverting etc. anytime you slide down the slope of quick fixes like employing fixed amount stops. you have just cheated yourself of the rewards that deep logical thought can provide. m

I know more than 14 methods of position sizing. But they are all based on money. I would create new position sizing techniques based on statistics (of trading system(s)).