2 ms·

Remember reading an article by an ex rentech guy who says that linear regression is about as complex as the modelling gets; that the edge is not in making thing

Remember reading an article by an ex rentech guy who says that linear regression is about as complex as the modelling gets; that the edge is not in making things hard.

XnoiVeX 8y ago
Deep neural nets approximate polynomial regression. That's probably the limit now.
- darksaints 8y ago
  The finance industry loves deep neural networks because it provides them a way to extract money from all of the naive tech engineers who took a deep learning class on Coursera and thought they could apply neural networks to trading. The general consensus is that deep learning for general trading purposes is complete bullshit because it unavoidably overfits. There are some quality uses for NN, but price or volatility prediction is not one of them.
logicchains 8y ago
I'd be surprised if that's the case, as surely somebody else would have replicated it by now. https://www.quora.com/What-are-the-investment-strategies-of-James-Simons-Renaissance-Technologies-I-understand-he-employs-complex-mathematical-models-along-with-statistical-analyses-to-predict-non-equilibrium-changes/answer/James-Baker-69 https://www.quora.com/What-are-the-investment-strategies-of-... describes something that sounds much more complex than what could be achieved by linear regression alone.
- usgroup 8y ago
  Tbh I imagine their edge is more likely to be better data and a secret theory. Then the applied part of it is simple but different: different analysis, different theory, different data, extraordinary results. Edit: just read your article ... kind of poo poos my theory if what he says is true ; at least at the level of the mega model.
  - saosebastiao 8y ago
    It doesnt refute your theory, it's just that your theory applies very specifically to HFT. Alpha generation in HFT is incredibly simple, and HFT analysis typically relies on very basic math: linear regression and recursive filters. HFT relies on execution far more than modeling to generate returns. But HFT is inherently low capacity. You can't put a billion dollars on an order book and expect to make the same returns as you would with a thousand. That's the reason HFT firms are almost always proprietary trading firms...they don't need or want more capital. A hedge fund, like Rentech, is typically on much higher timeframes because their size necessitates higher capacity strategies. This could mean holding periods of minutes for the million dollar funds to days for the billion dollar funds. As you get to higher and higher timeframes, your mathematics are going to need to be dramatically more sophisticated in order to beat the market. The math you are looking at in that quora link is about pairs trading which is where pairs of instruments mean revert over time. I would expect Rentech to be doing this type of trading over long timeframes and holding a trade for days to weeks.

Remember reading an article by an ex rentech guy who says that linear regression is about as complex as the modelling gets; that the edge is not in making things hard.

XnoiVeX 8y ago

Deep neural nets approximate polynomial regression. That's probably the limit now.

darksaints 8y ago
The finance industry loves deep neural networks because it provides them a way to extract money from all of the naive tech engineers who took a deep learning class on Coursera and thought they could apply neural networks to trading. The general consensus is that deep learning for general trading purposes is complete bullshit because it unavoidably overfits. There are some quality uses for NN, but price or volatility prediction is not one of them.

logicchains 8y ago

I'd be surprised if that's the case, as surely somebody else would have replicated it by now. https://www.quora.com/What-are-the-investment-strategies-of-James-Simons-Renaissance-Technologies-I-understand-he-employs-complex-mathematical-models-along-with-statistical-analyses-to-predict-non-equilibrium-changes/answer/James-Baker-69 https://www.quora.com/What-are-the-investment-strategies-of-... describes something that sounds much more complex than what could be achieved by linear regression alone.

usgroup 8y ago
Tbh I imagine their edge is more likely to be better data and a secret theory. Then the applied part of it is simple but different: different analysis, different theory, different data, extraordinary results. Edit: just read your article ... kind of poo poos my theory if what he says is true ; at least at the level of the mega model.
- saosebastiao 8y ago
  It doesnt refute your theory, it's just that your theory applies very specifically to HFT. Alpha generation in HFT is incredibly simple, and HFT analysis typically relies on very basic math: linear regression and recursive filters. HFT relies on execution far more than modeling to generate returns. But HFT is inherently low capacity. You can't put a billion dollars on an order book and expect to make the same returns as you would with a thousand. That's the reason HFT firms are almost always proprietary trading firms...they don't need or want more capital. A hedge fund, like Rentech, is typically on much higher timeframes because their size necessitates higher capacity strategies. This could mean holding periods of minutes for the million dollar funds to days for the billion dollar funds. As you get to higher and higher timeframes, your mathematics are going to need to be dramatically more sophisticated in order to beat the market. The math you are looking at in that quora link is about pairs trading which is where pairs of instruments mean revert over time. I would expect Rentech to be doing this type of trading over long timeframes and holding a trade for days to weeks.