Ernie Chan claims that statistical arbitrage is better than momentum like trend
- Thread starter GloriaBrown
- Start date
May be I am wrong but for me market wizard is a big book of marketing. The book itself helps the author to make money, and all those famous traders just talk whatever they want to present to cover up stuff or get more big customers. For me, may be I am wrong again, Paul wants to cover up his insider news trades by using the wave theory. He can use this to cover up whatever he wants with subjectively wave prediction expatiation even in court. Again, all of these may be wrong and I am just guessing. May be Paul is the real Jesus and use wave theory to 100% predict the market future, who knows.
(1) Statistical arbitrage is based on the premise that two assets' prices can be correlated, and that this correlation persists over time.
(2) Momentum trading is based on the premise that prices of a single asset can be auto-correlated (price tomorrow is correlated with price today), and that this auto-correlation persists over time.
Either premise can be true SOME of the time, but neither is true ALL of the time.
If market conditions are such that premise (1) is more true than (2), then Ernie Chan will be right. Else, he will be wrong.
What sort of conditions are holding these days? Very hard to tell without conducting some sort of empirical study.
Markets are living, breathing organisms. They are composed of real people who have free-will after all. So things change constantly. So it is stupid to make such blanket statements like one being better than the other. There are just too many variables involved to make fair comparisons.
BTW, don't be bummed about MATLAB being expensive. You don't need MATLAB for trading. Combination of Python + Numpy + Matplotlib is a much better choice (all open source).
Also, people make statistical arbitrage to be way too complicated with concepts of co-integration, ADF tests etc. Most of the advanced statistics is highly questionable since it is based on assumption of Normal distribution, which is a total nonsense in context of markets.
(2) Momentum trading is based on the premise that prices of a single asset can be auto-correlated (price tomorrow is correlated with price today), and that this auto-correlation persists over time.
Either premise can be true SOME of the time, but neither is true ALL of the time.
If market conditions are such that premise (1) is more true than (2), then Ernie Chan will be right. Else, he will be wrong.
What sort of conditions are holding these days? Very hard to tell without conducting some sort of empirical study.
Markets are living, breathing organisms. They are composed of real people who have free-will after all. So things change constantly. So it is stupid to make such blanket statements like one being better than the other. There are just too many variables involved to make fair comparisons.
BTW, don't be bummed about MATLAB being expensive. You don't need MATLAB for trading. Combination of Python + Numpy + Matplotlib is a much better choice (all open source).
Also, people make statistical arbitrage to be way too complicated with concepts of co-integration, ADF tests etc. Most of the advanced statistics is highly questionable since it is based on assumption of Normal distribution, which is a total nonsense in context of markets.
Quote from GloriaBrown:
My last post using Ernie Chan terms in the topic making it so confusing so I want to make this new post to talk about what I really mean. Ernie Chan has two books and he claims that statistical arbitrage strategies with two or more derivatives (he means the term Mean-reverting to describe these ) are much better than any kind of "momentum" strategies like trend following.
I don't know how to design a statistical arbitrage strategy to backtest to see which one is better. In his new book, he has like at least 6 math formula to find out a relationship between two derivatives before talking about strategies, and I still don't really get the real formula of these except a general concepts after reading like half of the book.
Except being heavily talking about matlab and very likey matlab sponsor him, he talks very professionally and he had a lot of working experience in statistical arbitrage trading desk in banks and funds. He is not like other "coaches" that just talks without any real prove of making money.
I am one of the active participates in his blog for months. I know if I ask him question like this, he would just reply what he said in his book, so I want to see how other think over here.
For me who doesn't live in US, using statistical arbitrage is something not easy even not consider how to implement because
1. I would need to access to many different derivatives and I have to do both long and short, which means I have to access to US markets since most of these are over there, while my time zone is exactly opposite to US, then I basically either stay awake every night. Even auto trading, I should stay awake to make sure everything goes well.
2. Double commission with at least two derivatives instead of one. In his book there are no commission charge.
3. I need to transfer my money to US dollars, then later transfer back to my own currency. I counted it and this is already at least a 0.2% cost.
Of course if statistical arbitrage is really much more powerful, then all these 3 factors are almost nothing.
So how do you guys think? I am still reading his book and just implement all of his math formula to test out a statistical arbitrage is already much harder than any momentum ways I know.
Please share your thought
stat arb is akin to trading spreads... and you answered your own question. the margins are usually better for spread trades than trading directional on the outright/underlying.. if relationship between two instruments exists one leg going crazy isn't going to really hurt ur position as the other leg will catch up.
the really successful traders I know in the real-world have made their wealth by trading spreads (not saying they don't trade directional). it's better but not as exciting
Thanks for sharing with insight. If normal distribution is not the best, then do you think we should use another kind of distribution or theory?
Quote from Eyez:
stat arb is akin to trading spreads... and you answered your own question. the margins are usually better for spread trades than trading directional on the outright/underlying.. if relationship between two instruments exists one leg going crazy isn't going to really hurt ur position as the other leg will catch up.
the really successful traders I know in the real-world have made their wealth by trading spreads (not saying they don't trade directional). it's better but not as exciting
May be you are right and I think this can be only known if you are in traders industry in ibanks/banks.
I think the disadvantage of stat arb for individual trader is that it is almost impossible to do that in day trade time frame because it would request high frequency level of speed to do so, then individual traders have to hold overnights with stat arb and this only factor can be so risky since we can have much better backtest stats with day trade but not overnight.
If not Normal distribution, then what?
Can't really give a definitive answer, since no one really knows. But you can look to history for clues.
As an example, options trading has existed in various forms long before the Merton-Black-Scholes model came into being. How did those options trader thrive? The answer is "heuristics". In fact, even today, options traders use some very sophisticated heuristics, and rarely apply the Black-Scholes equation.
Heuristics are rules of thumb that get the job done most of the time, but may fail some of the time. Successful trading will happen through good heuristics that have been validated through empirical data, not necessarily through the most elegant mathematical theory.
The good thing about mathematical theory is that it can be more easily automated. But I think a lot of people tend to go overboard with it. In the end, simple heuristics based on simple moving-average to gauge trend can be shown to work quite well compared to more complicated indicators.
A good heuristic for statistical arbitrage could be to just take pairs of highly correlated assets, and regress price of one agains the other over, say, a year, to get the ratios. Then use those ratios to construct a mean-reverting portfolio that can be traded rather mechanically.
You will still need to find additional heuristics to answer questions like which asset should be the independent variable in regression, how much of historical data to use for regression, how often to re-compute the ratios, what to do if the correlations start breaking down etc.
There are no clear cut answers here. Anything anyone says, I can easily play the devil's advocate against them. Some might say they have found the answer over years of trial and error and trading experience, but then that is just another way of saying that they have stumbled upon the right rules of thumbs (heuristics)!
BTW, don't be so hard on momentum trading. Just like there can be a real economic linkage to exploit in statistical arbitrage (like GLD vs GLX), there are some real economic reasons that give rise to trends and momentum (like a large institution being forced to break up a giant order into small chunks). If you say StatArb is better, then you are really saying that the latter economic reason is not as real as the former.
As I said before, both momentum trading and StatArb are betting on persistence of a statistical correlation, hence neither is a sure bet. Or rather, either can be a good bet SOME of the time.
(Edited for clarity).
Can't really give a definitive answer, since no one really knows. But you can look to history for clues.
As an example, options trading has existed in various forms long before the Merton-Black-Scholes model came into being. How did those options trader thrive? The answer is "heuristics". In fact, even today, options traders use some very sophisticated heuristics, and rarely apply the Black-Scholes equation.
Heuristics are rules of thumb that get the job done most of the time, but may fail some of the time. Successful trading will happen through good heuristics that have been validated through empirical data, not necessarily through the most elegant mathematical theory.
The good thing about mathematical theory is that it can be more easily automated. But I think a lot of people tend to go overboard with it. In the end, simple heuristics based on simple moving-average to gauge trend can be shown to work quite well compared to more complicated indicators.
A good heuristic for statistical arbitrage could be to just take pairs of highly correlated assets, and regress price of one agains the other over, say, a year, to get the ratios. Then use those ratios to construct a mean-reverting portfolio that can be traded rather mechanically.
You will still need to find additional heuristics to answer questions like which asset should be the independent variable in regression, how much of historical data to use for regression, how often to re-compute the ratios, what to do if the correlations start breaking down etc.
There are no clear cut answers here. Anything anyone says, I can easily play the devil's advocate against them. Some might say they have found the answer over years of trial and error and trading experience, but then that is just another way of saying that they have stumbled upon the right rules of thumbs (heuristics)!
BTW, don't be so hard on momentum trading. Just like there can be a real economic linkage to exploit in statistical arbitrage (like GLD vs GLX), there are some real economic reasons that give rise to trends and momentum (like a large institution being forced to break up a giant order into small chunks). If you say StatArb is better, then you are really saying that the latter economic reason is not as real as the former.
As I said before, both momentum trading and StatArb are betting on persistence of a statistical correlation, hence neither is a sure bet. Or rather, either can be a good bet SOME of the time.
(Edited for clarity).
Very impressive insight. Then which way do you choose to trade and why?
For me, I feel statistical arbitrage requires accurate math and total understanding to a pro level, if not big amount can be easily lost with hold two derivatives or more and don't know when and how to cut lose when the expected relationship is broken without noticed. Statistical arbitrage should provide more stable income but there is profit cap, unlimited potential lose when the market is stable without big accidentally crash kind of event.
For momentum ways, I feel it is more high risk high return way. Momentum cannot be as stable as statistical arbitrage but easier to control cut lose. Momentum can be easier and easy to tell if it works or not in shorter period of trading time.
statistical arbitrage requires 1. calculations to find a relationship between two or more derivatives, and have to make sure this relationship has strong math evidence or whatever proves relationship does exist 2. have a strategy to trade. Momentum doesn't really need 1. and easier to tell if 2. work or not
For day trade, momentum may be the only way left for individual trader as I explained above. I feel day trade is a safer way for individual trader with limited of money with much much much more data to backtest and know if live trades work comparing to overnight days/weeks/years trades.
For me, I feel statistical arbitrage requires accurate math and total understanding to a pro level, if not big amount can be easily lost with hold two derivatives or more and don't know when and how to cut lose when the expected relationship is broken without noticed. Statistical arbitrage should provide more stable income but there is profit cap, unlimited potential lose when the market is stable without big accidentally crash kind of event.
For momentum ways, I feel it is more high risk high return way. Momentum cannot be as stable as statistical arbitrage but easier to control cut lose. Momentum can be easier and easy to tell if it works or not in shorter period of trading time.
statistical arbitrage requires 1. calculations to find a relationship between two or more derivatives, and have to make sure this relationship has strong math evidence or whatever proves relationship does exist 2. have a strategy to trade. Momentum doesn't really need 1. and easier to tell if 2. work or not
For day trade, momentum may be the only way left for individual trader as I explained above. I feel day trade is a safer way for individual trader with limited of money with much much much more data to backtest and know if live trades work comparing to overnight days/weeks/years trades.