derail79
|
|
| Total Posts: 2 |
| Joined: Jul 2012 |
| |
|
Although I graduated from an ivy league school with masters in finance and have bachelor's in computer science I am by no means a professional quant. Based on my market action observations/gut feelings I've developed having traded my personal account for many years, I've coded two completely different stat arbitrage strategies, both showing ridiculous, too good to be true results in backtesting and trying to figure out where the bias is coming from.
Have been running one of the strategies live for a couple of weeks now. It made some money, but true level of robustness of the strategy is still inconclusive to me as more real time trading is needed. I back test it on 10 years of tick by tick data. I don't think the strategy is over-fitted to past results as it has only a few key parameters (degrees of freedom) and performs good (on paper) on different markets, instruments and timeframes. I know these backtesting results can't reflect a reality and I have no illusions about it, so please point me at the right direction on where to look for possible biases/wrong assumptions and most importantly on more reliable ways to evaluate performance of the system. Below are key figures of backtesting of one of the strategies. It places about 10 trades a day. All transaction costs are factored in.
Sharpe ratio 10
Profit factor 1.15
Maximal drawdown 7.45%
Profit trades (% of total) 59.31%
Loss trades (% of total) 40.69%
Average profit trade / loss trade 0.8
|
|
|
|
 |
Maggette
|
|
| Total Posts: 738 |
| Joined: Jun 2007 |
| |
|
| Did you test the strategy on a generated random time series? Not a test of your strategy, but for your backtesting procedure. |
Ich kam hierher und sah dich und deine Leute lächeln,
und sagte mir: Maggette, scheiss auf den small talk,
lass lieber deine Fäuste sprechen...
|
|
 |
IVolrev
|
|
| Total Posts: 76 |
| Joined: Oct 2007 |
| |
|
At the very first sight it looks like your average profit/loss is only a few pips, otherwise it would not explain you get such Sharpe ratios, given your profit/loss trade ratio is "only" 0.59/0.41 while your average trade actually loses money. It also strongly hints at a mean reversion strategy because your avg loss is larger than your avg gain. Thus, your strategy most likely turns over many trades per day in order to end most days in profitable territory. Am I correct? If not then I strongly assume you incorrectly calculate your Sharpe ratio (which is hopefully not the case with the academic credentials presented).
The Sharpe ratio actually should not be overrated in high frequency trading because if you do not end the day in profit most of the time your strategy is majorly flawed anyway. The max draw down looks to be in line, what would be interesting is if you could tell us what your back tested monthly returns are over the past year. I think only that additional piece of information will allow us to comment further. If it is in the region of 4-10% then what would be of importance is how long it takes to recover from your top decile drawdowns.
If you feel uncomfortable sharing those numbers then if I were you I would very closely investigate your execution model and whether you are correctly accounting for your costs of execution in your back tests. One approach is to monitor your average slippage in live trading and build in those numbers in your back test cost of execution model. Also, are your historical ticks correctly accounting for the spread you see in the data feed on which you trade? Hope to hear more,...
|
|
|
|
|
YukaRedux
|
| Now with added evil |
| Total Posts: 574 |
| Joined: Dec 2004 |
| |
|
| Do your live results of the past few weeks match the results of a backtest run over that same couple of weeks of data? That's where I'd start... |
The things you own end up owning you. |
|
|
j b
|
|
| Total Posts: 8 |
| Joined: Dec 2010 |
| |
|
| What kind of data are you using (trades/quotes/order book), and what assumptions are you making about fills? Are you assuming that you can buy at the bid and sell at the offer? |
|
|
|
|
|
Sharpe ratio 10
Profit factor 1.15
Maximal drawdown 7.45%
Profit trades (% of total) 59.31%
Loss trades (% of total) 40.69%
Average profit trade / loss trade 0.8
This system is "noise trading". This is evident from the low profit factor. As a result, the high Sharpe ratio is due to curve-fitting possibly.
Profit factor must be greater than 1.8, IMO, and stay close to that figure in out-of-sample testings.
Low payoff ratio of 0.8 is another indication of noise trading. This should be at least 1.
Usually these strategies make some money and then start losing fast.
"I don't think the strategy is over-fitted to past results as it has only a few key parameters"
Unfortunately, even 1 parameter will suffice for curve-fitting.
|
|
|
|
derail79
|
|
| Total Posts: 2 |
| Joined: Jul 2012 |
| |
|
| Thank you all very much for your inputs, I found the answer i was looking for. |
|
|
|
|
IVolrev
|
|
| Total Posts: 76 |
| Joined: Oct 2007 |
| |
|
| Dude, is this some sort of suck-em-up re-fill station? I mean nobody is asking you to share your strategy details. Yet, I think a minimum level of respect and politeness dictates to engage with those who volunteer to help you. Why would anyone engage with you ever again on this forum if you cannot even provide feedback what went wrong and how you fixed it. I at least won't ever respond to another question of yours unless you can over think your attitude a little. |
|
|
|
YukaRedux
|
| Now with added evil |
| Total Posts: 574 |
| Joined: Dec 2004 |
| |
|
I quite agree. That way blending lies, derail79...  |
The things you own end up owning you. |
|
|
|
Strange
|
|
| Total Posts: 1249 |
| Joined: Jun 2004 |
| |
|
| Actually, that is "The Way of The Internet", get a bunch of people do some work for you and don't even bother responding... but lets blend him anyway |
It's buy futures, sell futures, when there is no future! |
|
|
|
What do you think he fixed? The guy comes with a noise-fitted system with profit factor close to 1 and asks for advice. This means he is a newbie. He has no idea of what he is getting into. He is already inside the mouth of the monster called ruin. All we need to say to him is good luck. Do you think at his level he can provide any useful information? Do you really want to hear anything more from him? At least he thanked everyone. Others do not even do that. I think he was fine, a nice guy. He won't be trading for long though. Someone bought a genetic programming software and after 1000 generations he picked up the best curve. I told him don't do it. I told him that the guy who wrote the software is the king of selection bias. He didn't listen. His reply was that out-of-sample testing was good. Now he is 70% down. It took a few only for the sharks to take him to pieces.
|
|
|
|
|
mtsm
|
|
| Total Posts: 94 |
| Joined: Dec 2010 |
| |
|
wasn't that enough for you guys...?
"Although I graduated from an ivy league school with masters in finance and have bachelor's in computer science"
Wonder you even bothered replying? Maybe you went to ivy?
|
|
|
|
YukaRedux
|
| Now with added evil |
| Total Posts: 574 |
| Joined: Dec 2004 |
| |
|
| Why bother replying? Because we were all young and stupid at one point, and none of us got where we are today without some help along the way. And he asked nicely. |
The things you own end up owning you. |
|
|
|
mtsm
|
|
| Total Posts: 94 |
| Joined: Dec 2010 |
| |
|
no, I hear you, that makes sense.
I was simply saying that it struck me that the original poster felt the need to bring up his ivy league past in his first sentence. what does it have to do with anything?
I spent some long years in france and over there corporatism and nepotism are really really big, so I am somewhat sensitive... |
|
|
|
AIC
|
|
| Total Posts: 137 |
| Joined: Apr 2008 |
| |
|
Funny thread but perhaps I can still manage to get some goodwill out of all the kind contributors.
Wouldn't a lot of high frequency market makers (lets say equities) have systems in live trading that do indeed look like the original posters:
Sharpe ratio 10
Profit factor 1.15
Maximal drawdown 7.45%
Profit trades (% of total) 59.31%
Loss trades (% of total) 40.69%
Average profit trade / loss trade 0.8
Could it be they can capitalize on these so called "noise" systems due to their significant investment in technology and microstructure analysis? |
Train yourself to let go of everything you fear to lose- Master Yoda |
|
|
|
Scotty
|
|
| Total Posts: 661 |
| Joined: Jun 2004 |
| |
|
| In a sort of similar vein, are there any papers that cogently discuss the assessment of trading strategy results using these sort of terms? Or is it more like rules of thumb? |
“Whatever you do, or dream you can, begin it. Boldness has genius and power and magic in it.” |
|
|
