Sharpe ratio is a meaningless number as listed in the grid. We need a way to account for the real-life cost of trading.
SystemA: generate 30% annual returns by investing in AAPL in Jan 01 and selling Dec 31 with 20% std dev and risk free is 5%, then sharpe is (30-5)/20 = 1.25
SystemB: generate 30% annual returns by scalping system with 200 trades/year with 20% std dev and risk free is 5%, then sharpe is (30-5)/20 = 1.25
I choose A. SystemB has 200 commission fees * 2 for buy & sell that go to a brokerage. All other things being equal SystemA will excessively outperform SystemB while they have the same Sharpe ratio.
Until we figure out a way to subtract out trading costs, SQN and Sharpe both are meaningless metrics here.
Working on it.
This is really true of ALL stats. Drawdown, APD, Sharpe
1) But, don’t you already calculate commission?
- “after typical commission $142,421”
- "Realism Factor, with commis."
Just do your calcs after these are applied???. We really don’t need one algorithm for these, and another as applied to stats!!! C2 should be consistent across the board.
2) Additionally, I am struggling with these apparently multiple calcs/definitions:
—“Realism”
—“Keep after slippage”
—"Real trades from traders accounts/Autotrade results"
Are these 3 different ways to adjust for commissions+slippage ??? Why not harmonize or pick the best one? Again, be consistent. I really don’t see the purpose of all 3, They look like they all calculate the same principle.
And again, adjust stats after this is applied. It is much more realistic.
3) In summary, I would say that if all stats reflect true trading costs, that scalpers and people who depend on unreasonable fills would move toward better money management, fewer yet more cerebral trades, and would life the whole site.
There is a constant chatter about people who complain that their accounts do not reflect the system they trade. Adjust everything to reality, and respectability will jump.
4) Further, could someday, the “best systems / Grid / Old&Respec /” etc. reflect REAL trading - systems with good stats across the board? For example, when a top system has an APD of 0.07, it is almost untradeable long term.
I agree that is a problem. But as Ross says, this is as well true for the other statistics. This problem won’t go away by adding the proposed statistic.
BTW, I think your statement that they are ‘meaningless’ is too much black and white thinking. Between the black of meaningless and the white of perfectness there is a lot of grey. Although a high Sharpe (or any other statistic) in itself is not enough to recommend a system, a low Sharpe is enough to stay away from it. Subtracting commisions can only make it worse. Thus Sharpe can still be useful as a filter.
Index, all of those calcs you mentioned are measuring different things. Slippage has nothing to do with trading commission losses or leverage fees. People wanted to see real fills vs. theoretical fills and Matthew made a calc for that.
Jules, It is entirely possible in the world of databases, to set custom commissions, initial capital and leverage fee structures to each of us, based on our real-life brokerage accounts. And then calculate trading commission losses and margin fees on every system to get a “real annual return” rather than this purely academic, zero commission return that we see today. And then sort all annual returns and sharpe ratios on a post-fee basis. Agreed that I’m thinking black and white, I simply feel that the grid is so far from reality that it is not useful to anyone. The grid promotes unrealistic systems and degrades potentially good systems. I’d be happy to be in the gray, while right now we are purely in the black.
TJ, I won’t deny that it can be done. Also, I am definitely not against the idea to base all statistics on post-fee and post-slippage results. What I am saying is that even the present ‘best case’ Sharpe is useful to some extent, because you can elimininate the worst systems. A Sharpe based on post-fee and post-slippage statistics would be even more useful; agreed.
Let me add to this that I doubt whether it would make a big difference. Sure, there will be cases where the two ratios yield very different results. But taken over all systems I would be surprised if the rank correlation between the present Sharpe and the corrected Sharpe would be smaller than .90. Just an intuition.
The effect of commissions on returns is purely a function of initial capital, commission fees, and number of trades per year. Many of the C2 systems have excessive trade counts, some in excess of 500 trades/yr (1000 buy & sell). My broker charges $10/trade, so I’d have to pay $10,000/yr in commissions. Calculate net returns on Defiant with $10k initial capital and you’ll see their 200% pre-fee returns collapse to -100%. The commission effect is huge.
Regarding sharpe ratio, garbage in=garbage out. If we use pre-fee returns of +200% but our actual post-fee returns are -100%, we’ll find very different sharpe ratios. Fix the return calcs first, then worry about sharpe and other calcs.
