Neural Networks Hype

An excerpt from Andrew Lo:

A few years ago, I was contacted by a major pension plan sponsor. This is a sponsor of a pension plan of a Fortune 500 company. And, he called me up to ask whether or not I would be able to work on a consulting project having to do with asset allocation. And, I said, “Sure. I’ve done that before,” you know, “What are you looking to accomplish?” And, he said that, “We’d like to have you develop an asset allocation model for us using [neural?] networks.” And, I said, “Well, I do know something about that, but why in particular would you like to use [neural?] networks?” It was a rather strange request because this individual wasn’t particularly familiar with those sets of tools. And, he said, “Well, you know, we’ve heard that asset allocation problems are a very difficult and challenging one, and we don’t really have the staff to work on that. And so, what we’d like to do is to develop a system that essentially learns on its own so that, you know, you can develop it for us and then after a few years it will get good at it and we don’t need to worry about it anymore.” And, now, you laugh at that, but I mean, this was an absolutely serious request…

People love ‘black box’ nonsense.

---
You might like my 2017 epic release: Trend Following: How to Make a Fortune in Bull, Bear and Black Swan Markets (Fifth Edition). Revised and extended with twice as much content.

8 thoughts on “Neural Networks Hype

  1. But neural networks are supposed to be self-learning. The requestor probably just knew that and thought he could apply that to portfolio allocation.

  2. I have seen neural network software being sold on the Internet, prediciting daily closing prices. I thought it looked like a bunch of bs, but very well packaged bs.

    Now, they have developed computer software that can learn, and this software has been able to look at data (such as analyzing the swinging of a pendulum), and correctly ‘learned’ and derived one of Newton’s Laws of Motion based on it’s ‘observations.’ (I can forward the article if anyone wants it.) The problem, obviously, is that the laws of motion for a pendulu
    are fixed, and so reading data can lead to an accurate ‘understanding’ of the law. But as one person posted on you FB page, the market is not governed by rules in the same sense as they are in physics, and any related data points could be based on anything. As we all have read, seen, and heard several times, past performance does not guarantee future results. So even some neural net finding little hidden correlations still wouldn’t necessarily mean anything.

    Applying that kind of program misapplies the assumptions inherent in the program itself, and that is: given some or enough data, the ‘rules’ of the market can be deciphered. Total crap.

    Ps, linear regression doesn’t work for markets either, bc price data for financial instruments does not follow the bell curve. Another tool incorrectly used to ‘interpret’ market data. And it’s on a lot of company’s software packages.

  3. And I typed that last comment on my iPhone (regarding the typos) – still getting used to the keypad. I actually do know how to write!

  4. Michael: what do you think about using neural networks tools with trend following?
    Maybe we can identify more accurately when a stock is in trend and when maybe not.

  5. Mike – did you read that Andrew Lo paper on AMH?

    I would be interested to discuss what you think about it in light of systematic trading (in general) and Trend Following in particular.

  6. I’m not an expert, so this is only an idea.
    With a database in which we have data regarding stocks.
    For each stock we need to have the quotes and the ranges in which the stock is in up-trend or in down-trend.
    This is and hard work and that kind of trend analisys has to be done by an expert like you.
    With all this database we can teach a neural network system to detect when a stock has good probabilities to continue his trend in up or down market.
    What do you think about?

Comments are closed.