Why does my 60% win rate mean-reversion setup on gold still lose money? - page 2

New comment
Wasun Angkanapanit:

Is a 60% win rate with a 1:2 risk/reward just a mathematical illusion? I worked out that I'd need about 67% to break even before spread.

Suppose we win 1 ATR = 10.00 at Price = 100.00, our P/L is 10%.

However, when we lose, suppose 2 ATR = 20.00 at Price = 80.00, our P/L is -25%. It is not -20% so that we could recover by 2 wins.

Spread, commission and slippage make the result worse. 

When we risk 2 to get reward 1 and the number of times we get a reward doubles the number of times we get a loss, our strategy is still random in nature.

For #11: a move from 100 to 80 is a 20% loss measured from 100. The 25% is the gain needed to get from 80 back to 100. Dividing the loss by 80 changes the denominator.

For fixed cash outcomes of +1 per win and -2 per loss, two wins offset one loss before costs: p*1 - (1-p)*2 = 0 gives p = 2/3. If position sizes or ATR values change, the cash outcomes needn't stay at 1 and 2; use the realised average win and loss for that calculation.

Ryan L Johnson #:

Most likely, yes. On a relate note, CME 1OZ micro gold futures contracts have a fairly standard 0.25 spread. Technically, that spread is variable but it rarely deviates from 0.25 (while avoiding the daily close). The CME is a centralized exchange that all CME futures broker-dealers share, so everyone gets the same spread, price discovery, order execution priority, and real volume access.

As I noted in the strategy thread, the strategy is a basic version of the rather well known XARD strategy. The slow MA was actually an EMA (not an SMMA) in the original XARD strategy. I altered it to make trading more selective. If were going to change a parameter, I'd start with that one.

The only problem I can foresee with that is the fact that gold has been unusually rangebound recently. I fear that slowing the exit would only increase stopouts.

On another related note, I've found that a Renko chart having a Renko brick size of 5.00 is good proxy for the M15 timeframe on XAUUSD but much smoother than M15. By breaking all of the bars down into equal sizes, you can get more timely entries and exits based on brick close.

Here’s the comparison you suggested: the original XARD with an EMA 50 slow line (green) vs. your SMMA 50 version, both with your daily-close rule.

[attach: xard_equity.png]

  • Your SMMA modification clearly helped. Original EMA 50: PF 0.69, positive in 1 of 11 years, and slightly worse than the random-time benchmark. Your SMMA 50 version: PF 0.81, and it stays above the benchmark through 2016–2025. So the extra selectivity you added does show up in the data.
  • With the fail-safe stop, your version gets to PF 0.83.
  • In my data, both versions are still net negative on M15 with the EMA(13) exit, and the last few months (the range you mentioned) hit both hard.

Since spread doesn’t seem to be the difference (my broker’s spread is close to the CME micro’s 0.25), my best guess is that your live edge comes from something the mechanical rules don’t fully capture. Or your live period simply differs from this 10-year sample, which wouldn’t be unusual.

I’m still curious about the Renko idea: do you keep the brick at 5.00 regardless of gold’s price level?

Thanks for taking the time on this. It’s been one of the more useful exchanges I’ve had here.

Minh Tue Le #:

Suppose we win 1 ATR = 10.00 at Price = 100.00, our P/L is 10%.

However, when we lose, suppose 2 ATR = 20.00 at Price = 80.00, our P/L is -25%. It is not -20% so that we could recover by 2 wins.

Spread, commission and slippage make the result worse. 

When we risk 2 to get reward 1 and the number of times we get a reward doubles the number of times we get a loss, our strategy is still random in nature.

Agreed on the conclusion. With 2:1 risk to reward against you, winning twice as often as you lose only gets you to break-even before costs. After spread, commission and slippage it’s negative. That matches what I found: the 60–63% win rates were still below the roughly 67% needed.

One small note on the percentages, just so nobody reading gets confused: P/L per trade is measured from the entry price (or from the account), not from the exit price. A 20-point loss on an entry at 100 is −20%, not −25% (20/100, not 20/80). The asymmetry you’re describing (a −20% drop needs +25% to get back) is real, but it applies to account equity when position size compounds. With fixed position sizes, two +10 wins exactly offset one −20 loss.

Either way, the point stands: the TP/SL ratio can change the win rate, but it can’t create an edge. That has to come from the entry, and in my tests these mean-reversion entries didn’t beat a random direction.

Den Martin #:

For #11: a move from 100 to 80 is a 20% loss measured from 100. The 25% is the gain needed to get from 80 back to 100. Dividing the loss by 80 changes the denominator.

For fixed cash outcomes of +1 per win and -2 per loss, two wins offset one loss before costs: p*1 - (1-p)*2 = 0 gives p = 2/3. If position sizes or ATR values change, the cash outcomes needn't stay at 1 and 2; use the realised average win and loss for that calculation.

Thanks, that’s a clearer way to put it, and a good point about ATR. In these tests both the stop and the target were sized in ATR at entry, so the cash outcomes varied from trade to trade. That’s why I judged the results on the realised average P/L per trade rather than on the 2:1 ratio alone. It came out at roughly −$0.75 to −$1.00 per trade at 0.01 lot after spread, so the conclusion holds with the actual numbers too.
Wasun Angkanapanit #:
I’m still curious about the Renko idea: do you keep the brick at 5.00 regardless of gold’s price level?
Yes, that's 5 big points (ahead of the decimal) regardless of actual point size (behind the decimal).
12
New comment