Skip to content

Adaptive Portfolio Allocations

May 6, 2014

adaptI wrote a paper with a colleague- Jason Teed- for the NAAIM competition. The concept was to apply basic machine-learning algorithms to generate adaptive portfolio allocations using traditional inputs such as returns, volatility and correlations. In contrast to the seminal works on Adaptive Asset Allocation (Butler,Philbrick, Gordillo) which focused on creating allocations that adapted to changing historical inputs over time, our paper on Adaptive Portfolio Allocations (APA) focuses on how to  adaptively integrate these changing inputs versus using an established theoretical framework. The paper can be found here: Adaptive Portfolio Allocations Paper. A lot of other interesting papers were submitted to the NAAIM competition and the rest of them can be found here. The method of integration of these portfolio inputs by APA into a final set of portfolio weights is not theory or model driven like MPT, but instead is based upon learning how they interact to produce optimal portfolios from a sharpe ratio perspective. The results show that a traditional mean-variance/Markowitz/MPT framework under-performs this simple framework in terms of maximizing the sharpe ratio. The data further implies that traditional MPT makes far too many trades and takes on too many extreme positions as a function of how it is supposed to generate portfolio weights.  This occurs because the inputs- especially the returns- are very noisy and may also demonstrate non-linear or counter-intuitive relationships. In contrast, by learning how the inputs map historically to optimal portfolios at the asset level, the resulting allocations drift in a more stable manner over time.  This simple learning framework proposed can be substantially extended with a more elegant framework to produce superior results to those in the paper. The methodology for multi-asset portfolios was limited to an aggregation of pairwise portfolio allocations for purposes of simplicity for readers. The paper didn’t win (or even place for that matter), but like many contributions made on this blog it was designed to inspire new ideas rather than sell cookie-cutter solutions or sound too official or academic. At the end of the day there is no simple ABC recipe or trading system that can survive indefinitely in the ever-changing nature of the markets. There is no amount of rigor,simulation, or sophistication that is ever going to change that. As such, the hope was to provide insight into how to harness a truly adaptive approach for the challenging task of making portfolio allocations.

9 Comments leave one →
  1. Albert permalink
    May 6, 2014 3:35 pm

    I looked at the paper that won the contest and I was surprised how a simple simulation that has been used by traders for ages has passed as a novel approach that btw does not deal with data-mining bias and it is based on a gross assumption that the interval of parameters chosen is relevant to the problem or in other words it assumes some form of data dredging to start with. You either submitted your paper to a rigged contest or the contest promoters have no experience in evaluating papers. The first place winner is just a terrible paper that just conveys many misconceptions of its author who would be better off selling tires instead of developing trading systems. For your readers here you are always first place winner.

  2. May 9, 2014 7:51 am

    As one of the judges, I agree in one sense, i had Jason and David’s paper rated in my top three. The competition was very close, however. I believe that the other judges thought the topic of data mining bias was important and the discussion in the paper would help focus attention on it. Hard to say the competition was “riggerd”. The judges were all more than twenty year industry veterans. The papers were submitted to them without identification of the author. This is the sixth year of the competition with entries from seven countries, numerous states and faculty members from the University of Michigan, Columbia Business School and The Wharton School. Of course the primary purpose of the competition was to bring attention to active management as opposed to buy and hold investing, and to provide traders and the active management industry with ideas for developing new dynamic, risk managed investment vehicles. Hopefully you can find some other helpful papers in the twenty three available on the website.

    • Albert permalink
      May 9, 2014 2:04 pm

      AS one of the judges you should really feel a lot of regret that a simple parameter sensitivity simulation was promoted by your contest as a novel way of developing trading systems. Especially when it provides no rigorous justification and it is based on a gross assumption that selection of a rangle of values is relevant to the final objective. Possibly your veterans are experts in golf but not in trading system development. Data minig bias is an old and settled issue (See Aronson, White, etc.). The winner seems to have just discover it. It is good however that you admit that what you thought was a top paper ended up at the bottom of the list because of your experts.

      • david varadi permalink*
        May 9, 2014 2:36 pm

        First of all–thanks Albert for the kind words. In my post I was just trying to imply my own intent/style when approaching papers or blogs which is to inspire rather than to instruct or analyze in detail. i found a lot of the submissions interesting and well worth reading despite the occasional duds or flaws that may exist. I agree with Jerry that the breadth of submissions was impressive. I am also aware from talking to Jerry that the criteria of the judges in the contest is based upon several different factors besides just the novelty angle. Whether this criteria was applied consistently to all papers is perhaps an open question. That said, the attention to data-mining if anything else–even if the paper had serious deficiencies (which i agree)– is a good thing and I give credit to the judges for rewarding at least the spirit of good system development. At the end of the day I write because I enjoy presenting new ideas, and the contest itself is just a means of perhaps getting greater exposure so that is what counts.


  3. Albert permalink
    May 9, 2014 5:17 pm

    David, the author of the first winner does not understand what data-mining is. He considers only a very narrow and limited impact of data-mining arising from parameter optimization and makes unjustified claims that can result in better or worse performance depending on some arbitrary choice of a range of parameters that he admits is beyond the scope of his work and that begs the question. This is the kind of paper certainly does not worth a prize. That it was awarded first prize it is highly peculiar. Read the CUSMA blog:

    He makes the point:

    “And herein lies the main problem. The scan range will determine the median. If the range is too wide, the estimate will be too low and based on data that is essentially irrelevant because the trader would never actually pick that combination of parameters. If the range is too narrow, the entire exercise is pointless. But the author provides no way of picking the optimal range a priori (because no such method exists). And of course, as is mentioned in the paper, repeated applications of SPP with different ranges is problematic.”

    Essentially the winner author presented another way of developing systems that is subject to data mining bias, not a way that avoids it.

    Finally, it is up to you to open yourself whether you can tolerate abuse by any group of self-proclaimed veterans. I do not. Good luck to all.

    • david varadi permalink*
      May 10, 2014 12:09 pm

      Albert, I realize that the author may not have a good understanding of the topic. I agree that it does not deserve to even place having read many of the other papers. I also think that the longevity of experience does not always correlate to knowledge or understanding- though often times it does. That said, I think the competition is fun and helps to broaden awareness for those of us that seek to write about tactical investment strategies. I personally think that the crowd should vote rather than a select few judges since there are so many smart people out there– and it would be fun to get their honest reviews much like your own comments.

  4. Roman permalink
    May 14, 2014 4:55 am

    Hello David,

    Really interesting blog/papers, thanks for all your efforts !

    One quick question about NAAIM: why did you not “compete” with MinCorr/MinVar/Risk cluster parity ? To me, this is – broadly speaking – a related topic (i.e., related to the unstability of parameters used in classical portfolio algorithms) and would also deserve some focus.


    • david varadi permalink*
      May 16, 2014 4:59 pm

      hi roman, we did not present that application but have tested training the algorithm for those objective functions and they appear to do a good job of matching the performance and in some cases exceeding them depending on the objective function. this of course assumes the use of the method in the paper. however, since risk is more predictable the difference in performance was insignificant for the most part. a lot of possible applications though i agree.

  5. Danton permalink
    May 25, 2014 4:22 pm

    Hi, great paper, “Find the next day returns for the nearest matches and average them to find the forecast” in step 5 Appendix A should be deleted as it doesn’t apply in this case, right?

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google+ photo

You are commenting using your Google+ account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )


Connecting to %s

%d bloggers like this: