Starting an investment business

[merkur]

..

 

[gettoasty]

Another statistics program

Never used it but I heard it is easier once you know compared to say SAS

Not going to lie but having read this thread and the statistics influence makes me want to go back now and get a master/PhD in statistics. For god sake I did double major in it =\

 

[NameBack]

Think I just hit on a major improvement tonight! The way I had been building the ground truths that the machine learning algos were training on was biasing them towards a certain subset of strategies. This actually has some benefits, as for stocks who perform well using those strategies, the overall out-of-bag error is much lower for the strategies that are overrepresented in the ground truth.

However, there have been some equities that stubbornly refused to perform well with our existing system, and it was hard to hit on why. But I finally realized it had to do with this biasing issue, and instead I adjusted the biasing based on which strategies were performing well during the build period, and it seems to have fixed the problem!

Not only should this make the system more robust and versatile, but I'm also thinking that there could be an ongoing adjustment system that purposefully realigns the biasing in the ground truth based on which strategies have been performing well up to that point.

Specifically I was playing around with CTIC -- just because it has a beta of ~5 -- and in the first two weeks of 2009 using this refined strategy, it yields 205%. Hilarious. Prior to this, CTIC was a non-performer for us.

edit: I'll also have to play around with an unbiased random selection version and see what happens there. Probably lower overall returns but more consistency, I would guess.

 

[NameBack]

Eh, the "breakthrough" was a bust. No good way to predict which way to up-weight ahead of time. Ah well, the system as-is works great anyway.

Here's all of our backtesting up to this point:

 

[SkinnyGoomba]

How did you come about those particular companies?

 

[NameBack]



How did you come about those particular companies?

Several are ETFs we're familiar with already, and as to the individual companies, we got some recommendations for volatile stocks, and we also just browsed for a wide range of betas. XOM is our lowest at .6 or so, and CTIC is around 5.4, with most around 1.5-2.

That is to say, I guess, not a lot of thought went into picking the equities.

 

[SkinnyGoomba]

I don't know the history of all of those stocks, I do know XOM and DOW's history's pretty well, but how well does the program handle unexpected events like bankruptcy or something like an oil spill? Something outside of normal trending.

 

[NameBack]



I don't know the history of all of those stocks, I do know XOM and DOW's history's pretty well, but how well does the program handle unexpected events like bankruptcy or something like an oil spill? Something outside of normal trending.

Generally it handles event risk pretty well, but major singular events like an oil spill (which are much harder to see coming than a bankruptcy) are tricky.

The only real solution there is to make sure we're adequately diversified and to have some of our portfolio composed of ETFs that don't have the same level of event risk as individual stocks.

 

[CunningSmeagol]



It's no guarantee of future returns, but like I said, our backtesting is totally unbiased, so that is what we would have earned. I'll post a chart of our results so far in a bit.

How do you know you're not just being compensated for risk? You're supposed to get high returns with high-beta stocks. As an investor, I would be more interested in the beta of your fund, not just returns.

 

[SkinnyGoomba]

I don't know the history of all of those stocks, I do know XOM and DOW's history's pretty well, but how well does the program handle unexpected events like bankruptcy or something like an oil spill? Something outside of normal trending.

Generally it handles event risk pretty well, but major singular events like an oil spill (which are much harder to see coming than a bankruptcy) are tricky.

The only real solution there is to make sure we're adequately diversified and to have some of our portfolio composed of ETFs that don't have the same level of event risk as individual stocks.

I was actually a bit surprised that you would use any individual stocks for this. Was it just impossible to find those betas in ETF's?

Cs makes a good point, it's worth while to see what the risk adjusted return is to see if you are really offering and advantage or just most risk.

In my own portfolios I don't pay attention to beta, but for something like this It should be one of the indicators of how much risk you are offering, in my opinion.

 

[NameBack]



How do you know you're not just being compensated for risk? You're supposed to get high returns with high-beta stocks. As an investor, I would be more interested in the beta of your fund, not just returns.

Our fund has a beta of 0.47

 

[NameBack]



I was actually a bit surprised that you would use any individual stocks for this. Was it just impossible to find those betas in ETF's?
Cs makes a good point, it's worth while to see what the risk adjusted return is to see if you are really offering and advantage or just most risk.
In my own portfolios I don't pay attention to beta, but for something like this It should be one of the indicators of how much risk you are offering, in my opinion.

Yeah, basically we were just looking for more volatility without going into leveraged ETFs (though we did test a few of them just out of curiosity). It seems to work somewhat better on ETFs overall, but there are many stocks that it generates consistent high returns on.

Here's a chart of our returns from 2009 through 2011 using all the stocks and ETFs we've backtested:

 

[SkinnyGoomba]

Looks good, interested to see how it continues perform.

 

[CunningSmeagol]



Our fund has a beta of 0.47

That is stunning if true. Care to share what market inefficiency you're exploiting or is that proprietary (or already covered in the thread)?

The natural question is what expertise or insight does your team have that has gotten you into a trade that has eluded everybody else? I'm not really looking for an answer, just voicing a gut reaction.

 

[NameBack]



That is stunning if true. Care to share what market inefficiency you're exploiting or is that proprietary (or already covered in the thread)?
The natural question is what expertise or insight does your team have that has gotten you into a trade that has eluded everybody else? I'm not really looking for an answer, just voicing a gut reaction.

I'd say our fundamental conceit is pretty simple. Instead of trying to predict market movement, we try to accurately gauge which of a set of simple trading strategies is currently effective, and assume that said strategy will continue to be effective into the very near future. We take a computationally intensive task (predicting price movement), and reduce it to a simpler task (accurately assessing current performance). Everything else is built on that fundamental idea. I don't consider that proprietary.

My gut reaction as to why we did this where no one else is? Well, for one, there might be quant firms out there doing similar things -- it's hard to know for sure since it's all so proprietary. However, I think it is generally a very novel strategy, and I think our non-finance background is our edge. We weren't boxed in by pre-existing beliefs about where the good trades to make are, or what the limits of quantitative analysis are. Most quant firms treat the idea of machine learning directly influencing trades as a fairy tale. We didn't see any reason to take that point of view. I think that's what led to our success.