Building the TAA Model – Step 2


In this series I’m taking a more technical look at the Tactical Asset Allocation (TAA) model tested here and here. Unfamiliar with TAA? Click for a primer.

There are three steps to determining each month’s asset allocation. Our previous post covered the first, identifying which asset classes are in a bullish uptrend. Here in step #2 we’ll rank those uptrending asset classes and narrow the list to a core set to be traded.

Why Rank?

…or put another way, why not just trade all of the asset classes that are uptrending?

Two reasons:

First, I want to maintain a consistent number of positions in the portfolio (rather than jumping from say a couple one month to all of them the next, and back again). This will help minimize transaction costs (and hassle) and make the analysis in step #3 easier. At this moment (subject to change) the max positions in my model is four.

Second, I want to prevent the portfolio from being too heavily weighted in highly correlated assets. At this moment, the model considers eight asset classes: real estate, gold, commodities, US 10-year Treasuries, the US dollar index (long), and stock indexes for the world’s top 3 economies.

In today’s market, all the stock indexes PLUS real estate PLUS commodities are highly correlated to one another. If next month only this subset happens to make it into our portfolio, we lose all of the benefits of asset class diversification that should be a part of TAA.

When faced with either issue in a given month (too many asset classes or too many highly correlated asset classes), ranking helps us decide which asset classes to keep and which to cut.

Criteria for Ranking

So we’ve established that ranking is important, but how do we determine our ranks?

Faber doesn’t cover this specific topic, but he frequently talks about a “rotation system” where assets are ranked based on momentum by taking the average % change of each asset over the last 3, 6, and 12 months (highest average wins).

I’m using a similar momentum-based criteria for ranking with one big difference.

Returns are (as you’ve heard me repeat ad infinitum) an illusion…they’re just a function of risk. Much more important than returns are returns relative to volatility.

So in my momentum-based rank, I’m looking at which asset has exhibited the highest volatility-adjusted return over various periods (highest average wins). This prevents the portfolio from always drifting towards high-volatility asset classes.

So, to review…

In this second step, we take our list of uptrending asset classes from step #1 and rank them by volatility-adjusted return (momentum). If there are either (a) too many asset classes or (b) too many highly correlated asset classes, we use those ranks to know what to keep and what to cut.

I’m past my self-imposed word limit. Stayed tuned for part 3.

[Edit: click for a summary of all posts in this series on TAA] 

Happy Trading,

Geek note re: “highly correlated assets”: it’s important to determine which asset classes are highly correlated using only the data available at that moment in history because many have changed their stripes over time (read more). For example, in today’s market, equities and commodities are fairly well correlated, but for most of the market’s history they were very much independent. I thought I’d mention it because this is a potential source of accidental curve-fitting.

. . . . .

To stay up to date with what’s happening at the MarketSci Blog, we recommend subscribing to our RSS Feed or Email Feed.

8 Responses to “Building the TAA Model – Step 2”

  1. 1 Enright

    Do I understand this correctly when I ask, is TAA really just feeding bubbles? What happens when all the TAA users shift from one momentum to another?

    • 2 MarketSci

      Hello Enright – sure, you could say that about any strategy based on trend-following or momentum. michael

  2. 3 Dennis

    Given that TAA is using ETF’s as the underlying vehicle, doesn’t this limit your historical data for backtesting , or are you using the index that these ETFs are supposed to track ?

    On a somewhat unrelated note, I’m curious what your thoughts are on gaining more leverage using options (for those underlying markets with option liquidity). You could negate some of the theta decay by either purchasing a few months out from the front month, and still accumulate more deltas as well as better defined risk, or put on a simple spread with additional calls to adjust your leverage / deltas. If the following month, your strategy dictates you continue to hold, simple roll another month out.

    Thanks –

    • 4 MarketSci

      Hello Dennis – for each asset class I’m using the underlying index until ETF data becomes available and then switching over to the ETF data.

      No opinion on trading the strategy with options (outside the scope of what I do).


  3. 5 snorlax

    Recently I had been doing some reading up on commodities and read a paper that you might find helpful for further insight. The paper is “Momentum Strategies in Commodity Futures Markets” by Miffre and Rallis from 2006, and is on SSRN. They set up an examination looking at basing returns on 1-3-12-24 months etc and then holding for 1-3-12-24, which unless I’m miss reading would be similar to your back-testing. They also have an updated version that I haven’t got around to that adds term structure and other correlatives.

    Appreciate the always insightful posts.

  4. 6 Darklingg

    Hi Michael

    Very interesting!

    How do you compute the volatility?

  5. 7 Michael

    How do you use correlation to reduce the list of asset classes in a bullish uptrend?

    In the example you site above (stock indexes PLUS real estate PLUS commodities) if they were all highly correlated and the ones selected for, which would you pick?

    If there were three highly-correlated assets and one uncorrelated picked, would you restrict the list to the top of the correlated and the non-correlated?

    Thanks for laying this out, very interseting.

    • 8 MarketSci

      Hello Michael – good question – you got it when you wrote:

      “If there were three highly-correlated assets and one uncorrelated picked, would you restrict the list to the top of the correlated and the non-correlated?”

      That’s exactly what I’m doing. Just one side note however: I have to be pretty conservative with how I define “highly correlated assets”.

      The market goes through periods (especially in today’s market) where almost everything is pretty well correlated, so we can’t set the bar for determining correlated assets too low. That means a stock index might make it in which real estate and commodities in the same month (in fact I’m expecting it to happen in our next set of trades).

      Not optimal I know, but it is what it is. I think in these months the model becomes more of a traditional trend-following strategy than gaining a significant advantage from diversification.


Leave a Reply

Please log in using one of these methods to post your comment: Logo

You are commenting using your 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 )

Google+ photo

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

Connecting to %s