Oh, the conundrum. I have the current version of Pairs Trading QID QLD Scalper, and I have a version that I've just named PTQQS 2.0 on the right.
PTQQS, the original, uses long and short exit thresholds that are very close to the long and short entry thresholds such that usually within 1 to 2 days we crossover the exit thresholds and sell on the open the next day.
What I did with PTQQS 2.0 was tighten the stop to be any close below a 0.67% loss to attempt to sell at that stop price. I made the exit thresholds nearly on the opposite ends of the fair value ranges. In order from least to highest value, you have long entry thresholds, long exit thresholds, both negative values, and short exit thresholds, and short entry thresholds, both positive values. Essentially what I've done is make the long exit threshold at the short entry threshold level, and the short entry threshold at the long entry threshold. This has allowed trades to run a lot more, and losess to be cut quickly. It is only in the current environment YTD that this version called PTQQS 2.0 outperformed the original PTQQS, which you can observe annually in the spreadsheet as well as the equity curve.
Long + Short Long + Short
Starting Capital $100,000.00 $100,000.00
Ending Capital <b>$705,222.89 $528,048.83</b>
Net Profit $605,222.89 $428,048.83
Net Profit % 605.22% 428.05%
Annualized Gain % 81.58% 66.23%
Exposure 24.82% 25.73%
The main catch is that the net profit from PTQQS is a lot more than PTQQS 2.0. But I have a bigger point to make below.
Number of Trades 80 64
Avg Profit/Loss $7,565.29 $6,688.26
Avg Bars Held 2.66 3.17
Winning Trades 61 38
Winning % 76.25% 59.38%
Gross Profit $1,006,072.77 $579,589.48
Largest Winning Trades $86,930.39 $64,212.48
Avg Profit $16,493.00 $15,252.35
Avg Bars Held 2.8 4.32
Max Consecutive 10 5
Losing Trades 19 26
Losing % 23.75% 40.63%
Gross Loss ($400,849.87) ($151,540.65)
Largest Losing Trade ($45,508.73) ($31,620.12)
Avg Loss ($21,097.36) ($5,828.49)
Avg Bars Held 2.21 1.5
Max Consecutive 3 4
Max Drawdown ($123,590.38) ($48,278.50)
Max Drawdown Date 8/17/2009 8/17/2009
Max Drawdown % -16.66% -12.32%
Max Drawdown % Date 8/17/2009 11/20/2008
<b>APD 0.7649 1.0745 </b>
This is the really big issue I'm struggling with, and why I've pointed out to Ross (Index) Trader Zones, that the Sharpe ratio also measures risk adjusted returns, and, in my opinion, has a more solid, theoretical base.
APAD 1.6675 2.4503
Wealth-Lab Score 273.8864 225.6714
RAR 328.6462 257.3874
MAR 4.8962 5.3745
Profit Factor 2.5098 3.8246
Recovery Factor 4.897 8.8662
Sharpe Ratio 2.0727 1.9084
In the case of these systems, I see one that has a much higher sharpe ratio, and the other one has a much higher APD.
Sortino Ratio 9.1023 5.2494
I don't normally use the Sortino ratio, but we can also see PTQQS, the original, is higher on a risk adjusted basis here with the Sortino ratio, too.
Ulcer Index 4.7827 4.5253
WL Error Term 6.459 6.001
WL Reward Ratio 12.6307 11.0359
Luck Coefficient 5.2707 4.21
Pessimistic Rate of Return 1.7801 2.6788
Equity Drop Ratio 0.0395 0.0363
K-Ratio 0.7742 0.6087
Seykota Lake Ratio 0.0474 0.0411
<b>Expectancy 0.6165 1.203
Expectancy Score 14.7419 22.5746
</b>
Not sure, don't really use this ratio, but it has the same conflicting signal.
Max Losers Held 1 1
Max Winners Held 1 1
So, I'm actually hoping Trader Zones could come debate this as to which one he would trade. I'm still of the mind I would trade the original because it has a higher Sharpe Ratio, as well as hundreds of thousands of dollars more profit.
I think it's a perfect opportunity to debate the pro's and cons of the APD statistic, here.
For those who aren't familiar with C2, there is a controversial statistic on the site that was developed in part to detect "averaging down", or as I understand it, letting losses run. It is found by the summation of total drawdown on each trade, a stat not even provided by mutual fund companies, by total net profit. An APD of 0.1 implies you risked $1000 for every $100 of profit. Mainly, the problem I have with it is its assumption that you sell at the lowest price on every trade. (Not a valid assumption). This statistic called the APD I wrote in wealth script code and apply it as a perf script in this excercise. Anyone using WL4.5 has access to the code in the knowledge base.
What I'm finding with the APD is that it might not be capturing risk adjusted returns for single entry, single exit systems properly. It was predominantly exposing averaging down systems that would enter trades after previous entries at lower prices on buys and higher prices on sells.
As we use it here, I found conflicting information. I have one statistic that points to using PTQQS 2.0 over PTQQS DESPITE the obvious difference in profitability. The Sharpe ratio gives a much different opinion on the system. Rather, it seems very obvious what the better system is, yet Trader Zones, the creator of the APD statistic would have you believe PTQQS 2.0 is better, despite the Sharpe and Sortino Ratios being better.
This year PTQQS 2.0 is up 30% more than PTQQS, but in previous years PTQQS 2.0 does not even compare to PTQQS.
So, if we have PTQQS 2.0's APD stat as 1.0745 to PTQQS's APD stat of .7649, what I must conclude is that the additional risk of (1.0745-.7649)*1000=$309.6/trade, is justified given that for every $309.6 extra risked dollars, I get (705222.89-528048.83)/(80-64)=$177,174.06/16=$11073.37/($309.6*16)=2.2354187 times my dollars risked extra over the APD stat, ceteris paribus.
I would have to say the extra profit is more than enough to cover the additional risk of adding extra trades. The other facet of this is how I was able to realize that if you buy something at 10 and sell at 11, buy back seconds later at 11 after closing the trade, and sell at 10, then all of that drawdown gets counted towards your total. It is in this example that I think the APD stat is a bit flawed, and could be improved by modification for cases like these.
It would seem to me, that, in this situation, it is not actually a new trade being entered, more than it is about a change of basis. If we did as the example, buy @10, sell @11, then buy @11, sell@10, I postulate the effect on the real APD statistic should be zero. Because the actual basis of profit in the security is still 10. So if you buy at 10, sell at 11, and then buy it quickly back at 11, and regrettably sell at 10 over any timeframe, what you have is not actually drawdown more than it is a change of basis. I know this is complex, but I think there's a theoretical basis to changing the way the APD stat measures drawdown. In this example, the drawdown would not change.
Another example that was given recently was if you had a system that sold for $110 profit per trade and lost $100 per trade, if you did that enough times your APD would be horrible, but your calmar ratio would be larger and enticing at that.
