Martingale debunked by Gary

[danielv]

2x nonmartingale or more is the best then. I have no point in arguing martingale or no martingale, I do care that we are looking for the same solution which is to increase profit and minimize drawdown. I guess I'll focus on finding method and setup that has high probability from now on. Anyway thanks for the excel file, I have downloaded it and experiment with it for a bit and come to the same conclusion like you.

[AllanS]

a 10-loss series results in a bet size of 2^10 = 1024 TIMES your initial bet size. If you started out with 0.01 lot, you'd be risking 10.24 full lots. All to win a 0.01-lot-sized profit. Does that sound like a sensible Risk:Reward to you?

What happens to the Martingale curve if winners are allowed to run?

As you say, it's crazy closing a $1024 trade for $1 return. What if you close the $1024 trade only when it returns $1024 (after covering accumulated losses)? ie. close the winner when it moves 2R. If random, 50% of 1R winners will go on to 2R. By waiting, you will magically change a $1 return into $1024, 50% of the time.

What if you only cover costs and let the rest run?

The other question is: what's the chance of blowing a M account before doubling the initial stake? Once doubled, you can trade risk free using foreign capital only. I wonder if Martingale actually gets you into risk-free trading faster and more reliably than the usual 2% MM etc.

[danielv]

Can you give the pips distance simulation AllanS ? I assume that each time the stop loss hit, and you double the lotsize, you also double the take profit distance. The practical implication is each time we lose, we only enter a trade that has double reward to risk ratio. Hmmm

[AllanS]

Can you give the pips distance simulation AllanS ? I assume that each time the stop loss hit, and you double the lotsize, you also double the take profit distance. The practical implication is each time we lose, we only enter a trade that has double reward to risk ratio. Hmmm

Hi Daniel,

Suppose the stop is 10 pips and the target is 20.

Trade 1 loses $1. Trade 2 loses $2. Trade 3 loses $4. Trade 4 loses $8.

Trade 5 goes up 20 pips and makes 2 x $16. This covers the $15 accumulated loss, leaving $17 profit. ie. I risk $16 to make $17.

But why cut the trade at 2R? Why not use Martingale to bring positions to breakeven (after covering accumulated losses), and let the remainder run? ie. Graham's millipede (Forex Factory).

Another thing. As more successful positions are added, foreign capital grows exponentially, providing a partial buffer against the exponential expansion of the Martingale.

[garyfritz]

Load up the Martingale.xls spreadsheet from post 1. Set the Win$ in cell B6 to 2x the Loss$ value in B7. That has the effect of setting the PT 2x larger than your SL.

With 40% wins (cell B5), and bailing if you take 5 losses in a row (B8), the Martingale is roughly comparable to trading a non-Martingale strategy with about 3x normal size. (I.e. instead of a Mart that starts out with 0.1 lot, then 0.2, etc -- just trade a non-Mart with a fixed position size of 0.3 lot.) You generally get similar returns, you do it with a known fixed trade risk, and you don't have to take positions 16x your original bet size.

With random entries, when TP = SL you will hit about 50% wins. With TP 2x larger than SL, random entries will hit about 33% wins. 33% wins with 2:1 TP:SL is basically breakeven, just like 1:1 TP:SL with 50% wins. So you still need a winning strategy. The Martingale won't turn a breakeven (or losing) strategy into a winner. Sometimes it makes more than the non-Mart breakeven system, sometimes less, but overall it's about the same. Just riskier.

[AllanS]

Hi Gary,

I believe runs are not random, but driven. Any strategy that guarantees you're on the right side of the market when a run occurs will automatically benefit from this edge. I'm wondering if Martingale can be used to establish multiple, small B.E. positions with the goal of letting them run.

I agree there's no edge in Martingale itself, or in any MM strategy. Hanover (I think) presented a proof to that end in FF. But random Martingale combined with non-random runs may well have an edge.

BTW, loved that excel model. Kept me happy for an hour or two. Reminded me once again of how wildly results can vary, driven by chance alone.

[garyfritz]

If you have more than 33% wins with 2:1 TP:SL, your model is not random. You will make a profit with your Martingale, or any sensible MM scheme.

BUT the key point I keep hammering on is -- even with a non-random strategy, you can get similar returns with MUCH less risk and smaller position sizes if you just trade a fix position on all trades. With 2:1 and 40% wins, running a 3-unit position gets you roughly as much profit as a 1/2/4/8/16/bail Martingale.

And with intelligent sizing, you'll make a lot MORE, because you have to size the Martingale positions so 1+2+4+8+16 = 31 units is your appropriate risk size -- e.g. if you're risking 1% per trade, you need to size your positions so 31 units is a 1% risk. Whereas with fixed 3-unit trades, you can have 10x larger positions, so in reality you'd make WAY more profit than with the Martingale.

Probably a single-unit position size, sized to 1% risk, will return a lot more profit than a Martingale starting out with 1 unit and sized for 31-unit = 1%.

[AllanS]

Probably a single-unit position size, sized to 1% risk, will return a lot more profit than a Martingale starting out with 1 unit and sized for 31-unit = 1%.

Yes. Using M to find tops and bottoms can work spectacularly when the successful position is loaded after several doublings. But it's very annoying to find a major turning point with a position of only 0.01 lots. A larger, fixed position would smooth this out.

Back to your excel model. With a 55% hit rate, a position of 5% had about a 1% chance of going bust before doubling the initial stake. A 1% position didn't even hint at going bust after several hundred trials. Again, I was impressed that some accounts grew to 30000, while others dithered around 10000, all with the same initial odds. If trading the former, you'd think you were a genius. If the latter, you'd despair, even though in fact you'd be precisely as good as the other bloke! A tough game, this.

[garyfritz]

55% wins with what W/L ratio? Assuming W = 100, L = 100, then the AverageTrade = 0.55*100 - 0.45*100 = 10. The Kelly value for that system is AverageTrade / AverageWin = 10 / 100 = 0.10. Meaning that risking 10% per trade would maximize profits over the long haul, but at the cost of 90% DDs or worse.

So your 5% risk was risking 1/2 Kelly. That's extremely aggressive. I'm a lot more comfortable around 1/10 Kelly, which is what your 1% risk was.

BTW you should be aware: that simulation XLS uses fixed position sizes. So if you specify Win$ = 100 and Loss$ = 100, then you will always win or lose $100 on each single-unit trade. So it does NOT reduce its position size if you go into a drawdown. The chances of it blowing up its initial $10k account value are much higher than if you risked a fixed percent of your account instead of a fixed dollar amount.

How did you do the 1% vs. 5% position size test? Did you modify the spreadsheet?

[AllanS]

Hi Gary,

I set the gain to 500, loss to -500, %win to 55, and hit F9 200 times. It went bust (before hitting 20000) three times. Not a big sample, but hey, my coffee was getting cold.

Making the losses 520 and the gains 480 (to simulate slippage and spread) increases the risk of ruin to around 12%.