System quality number
System quality number
This was put into EWFO at the request of a client.
My question to this client was, did SQN work well compared to net profit / draw down or net profit * average trade.
"Absolutely YES !!!
At least with my own trading systems, I am experimenting that the table of SQN value is true and when there are setting having, at the same time minSQN > 2.5 and Rank > 15 the stability is really good and OOS data are very close with IS. On the other side using P/DD only (with the same trading system) it is more frequent to get better IS but less robust and then worst OOS.
The results are extremely accurate when the trading system has time filter and or day of week or other seasonality filter, in this case the core edge of market flow is exactly to generate very consistent trade statistics and then using SQN you should easily achieve values > 3.5 and very high stability."
Below some information about SQN (System Quality Number).
- SQN is a trading system metric developed by Van Tharp http://www.vantharp.com/tharpconcepts/sqn.asp
- Mr. Vantharp experimented that higher SQN correspond to higher probability of a trading system to achieve profit goals with proper money management
- Here there is a trading system ranking of SQN meaning
- SQN measure how much the statistic of a trading system trades is "stable", higher value means lower probability that the results are depending on random lucky sequence and higher probability that the trading system captured a real market edge. For this reason using SQN in WFO metric decrease probability to curve fitting and then increase the forecast capability of the whole WFO process and, according with EWFO measurements, typically provide better stability score.
On top of this I wanted to:
- mix SQN with Profit/DD in order to capture setting that at the same time are good (higher P/DD) and stable (higher SQN)
- measure P/DD and SQN on 3 different time segment of the available history and use the lower. This because the "weakness of chain is the weak of the weaker ring", then doing so we are even more safe to catch the most robust trading system setting.
- the final "Rank" metric is the combination of the "worst P/DD" and the "Worst SQN" and I believe is a strong metric to keep out curve fitting and catch good and stable setting.
My experience using it in EWFO is that the result of using Rank is quite equivalent to use minPDD and minSQN with the same weight, anyway I decided to systematically use Rank, minPDD and minSQN with the same weight as my own standard target metric.
The inner SQN workings.
Let's assume on each test we have N Trades, where 1 is the most recent and N is the oldest (like easylanguage convention)
Split on 3 different groups:
Group 1: 1 to N
Group 2: 1 to N*0.67
Group 3: 1 to N*0.33
NP/DD1 NP/DD on Group1 (then equal to NP/DD definition)
NP/DD2 as NP/DD on Group2
NP/DD3 as NP/DD on Group3
SQN1 as SQN on Group1 (then equal to SQN definition)
SQN2 as SQN on Group2
SQN3 as SQN on Group3
MinNN/DD = Minimum between NP/DD1, NP/DD2, NP/DD3
MinSQN = Minimum between SQN1, SQN2, SQN3
And then the targeted functions to be added are:
5) Rank = MinNP/DD * MinSQN
Obviously all 5 should work properly on each time interval IS and OOS.