Recently began speaking to a rather impressive individual with an MFE from Stanford about an algo he has developed for Forex trading. My initial analysis in regards to the descriptive statistics has found it to not only be profitable on average, but also, generating returns that are statistically significant and different from zero.
My main question is, what are some other criteria I need to consider in terms of determining whether what he has built is actually a viable option for achieving alpha? I think although I have quite a large sample to draw from (thousands of trades since Jan 1st, 2017), the time period (4 months) seems pretty short when years of consistent performance would be ideal.
Also, what is a way I can verify these returns are legitimate? Is there an auditing process I could employ? Std Dev. of returns seems pretty high but monthly returns have average ~6% which when annualized are around 74% (Obviously dangerous to assume).
I do not know much about the more technologically driven aspects of machine learning when it comes to finance so any input would be appreciated.
Tried that. He isn’t interested in selling it. Rather wants a cuts of profits. He is in the UK and says that I would be the only US based investor with any rights. Not sure how I could monitor any of that though.
Ha, isn’t this the million dollar question. What time period was the algo developed over? I’ve listened to some recent talks on how to think about this issue, but I’m not a quant so most of it goes over my head. I just know the common sense questions to ask
First of all, Stanford has no MFE program. You probably mean the MS in Financial Math program. Second, you should know that 60% of people who graduate from this program claim to have some money making algo that has unbelievable Sharpe ratio and that is going to make a lot of money in the real world, despite having only hypothetical returns. I’d be extremely skeptical of committing capital to ideas like this without completely understanding how they work.
OP, the main concern I have is the time frame of the data. Even though thousands of trades are plenty judge an algo’s performance and stability, 4 months is not enough time to see how it will perform in various market conditions. If this thing is HFT it will be highly dependent on market liquidity. Time of the year, and/or in times of greater volatility, the structure of market participation and liquidity changes significantly. Will this algo be able to negotiate that change?
I agree. I thought about regressing returns against time of year since I really do not know a ton about the underlying assumption of the algo, to see if there is any seasonality. However, I do not think I can do much meaningful analysis in that regard with 4 months of history.
Yes, that is the program. The sharpes for what little data I have are mixed. Very low for the month of April, thus far, but then almost double the result for January/February. The returns though, are realized returns from what he has given me, but I am skeptical (although I guess I shouldn’t be since it is only 4 months).
I honestly only plan to maybe commit a couple grand max to start. The conversations hasn’t been too in-depth yet on how it actually works but I am in the same camp that I need to understand it more and have him go into more granular detail. Although, not sure how much he will disclose without feeling like he is giving me proprietary insight that I could replicate or tell someone else about.
This is red flag enough imo, any truly worthwhile HFT will have stable returns.
Statistical analysis of returns isn’t the answer. He should have a structural reason why he makes money, ask him infrastructure questions, don’t look at a time series of returns. M2c.
The returns are stable. 5 percent, 6 percent, 7 percent, 6 percent, respectively since Jan 2017. You mean stable in terms of being adjusted for risk? It is the std dev. that is all over the map. I definitely need to ask him some pointed questions, agreed.
Yeah, I talked to him more tonight and he contradicted himself a couple times when referencing some basic fundamental FX concepts so that was concerning. Plus, if something bad happened with it between now and sitting for L2, I would be extremely pissed.
Any edge in electronic market can only be achieved via something like building radio towers to speed up communication between Mahwah datacenters and chicago… (true story). If you think some one has developed ‘revolutionary’, apha generating TRADING strategy, i would be VERY skeptical
Backtesting - what did that show? over what periods? which periods were out-of-sample? etc. Where did he get his data? Slippage? Spread accounted for?
Scaling - If it works with 10,000 contracts, how about 2m contracts, how does that affect execution? Did he take that into consideration in the backtest?
Also, make sure the guy isn’t martingaling - that is, just doubling up losing trades until they are winners or he goes bust. That works for a while until it doesn’t. BOOM.
I wouldn’t touch anything like this with a 10 foot pole until it has years of success under it’s belt. If it’s so good, why doesn’t the guy just run it himself with his own money and make himself rich and retire? That’s the million dollar question right there.
i mean if its early on and the guy doesnt have much capital of his own to invest you could see how they would want to get some cap intro. That being said this thing screams scam and I wouldnt get near it either.
