We’re very excited for the launch of an awesome new feature for our members: Meta Strategy.
We track a wide range of published tactical asset allocation strategies in near real-time (40 and counting), which members can then combine into their own custom portfolios. Our platform helps members better understand how each strategy fits into a coherent trading plan, but we’ve never given members specific guidance on just how to combine those strategies. That’s not going to change. We realize that the “best” approach is unique to each investor. But we also realize that the sheer volume of data available on our site can be a little like trying to take a sip of water from a firehose.
So in response, we’ve launched Meta Strategy, our own smart approach to combining the TAA strategies on our site. Members can follow Meta in near real-time just like they would any other strategy. In this post we’ll describe the logic behind Meta in more detail. First, let’s look at results since 1973, net of transaction costs (learn more backtest assumptions).
Linearly-scaled. Click for logarithmically-scaled results.
The Importance of Walking Our Test Forward
We don’t call Meta Strategy a “smart” approach to combining strategies just because the backtest is sexy. It is sexy, but that’s the easy part. Given such a large pool of strong strategies to choose from, we could slice and dice them into the perfect configuration to produce even better historical results than those we’ve shown here. But that would say nothing about whether that approach was best suited for the future. We can’t eat backtested returns. All that matters is how we perform out of sample.
So all of the testing shown above is “walked forward”, meaning at each data point Meta only had access to the data that was available at that point in time. Meta can’t “peek” into the future the way we could if we simply chose the best performing strategies as of today. That provides a certain degree of confidence that our approach is targeting future outperformance, not simply past outperformance.
A Representative Sample of “Good” Strategies
Each month, Meta selects 10 TAA strategies based on a number of factors that we’ll discuss in a moment. It’s important to understand that Meta is not “timing” these strategies. In other words, we’re not aggressively moving in and out of one or a few strategies, trying to identify the very best.
There are two reasons for this. First, predicting which single strategy will be the next top performer with certainty is impossible. And second, adding aggressive timing techniques (Meta) on top of aggressive timing techniques (the individual strategies) would increase the opportunity for overfitting to history and the likelihood of disappointing results out of sample.
So instead, Meta selects a representative sample of “good” strategies. It separates some of the obvious chaff from the wheat, but it doesn’t attempt to pick the absolute best. Diversification across good but different strategies has the same benefit of diversification across good but different assets.
Our Approach to Selecting Strategies
First, we measure the performance of each strategy relative to volatility, drawdown and other measures of risk across multiple timeframes. Returns in isolation are arbitrary. Returns over a single period are arbitrary. We’re looking for consistent performance relative to risk.
Most of the strategies that we track trade monthly at the close on the last trading day of the month. For these monthly trading strategies, we measure performance across all potential trading days, rather than simply month end. We’ve talked a lot in the past about alternate trading days and how they can be useful to ferret out potential overfitting (read more). By measuring performance across all potential trading days, we’re capturing a truer view of each strategy’s performance.
We adjust our measurements for the similarity between strategies. We understand that there isn’t a lot of difference in terms of expected future performance “at the margins”. For example, knowing what we know today, we can’t say whether the 11th best strategy will perform better than the 10th best. So we put a premium on strategies that are more dissimilar to their counterparts.
We also discount each strategy’s performance for the impact of current interest rates across the yield curve. It’s impossible to say with any certainty whether stocks, bonds or any other asset class are going up or down in the next month. But we can say whether the returns for certain interest rate sensitive asset classes (like US Treasuries) enjoyed a tailwind in the past due to consistently falling interest rates over the last four decades. Given how low rates stand today, there are simple mathematical limitations on how well these assets can perform moving forward (read more).
Each month, we select the top 10 strategies based on the criteria above. On average, Meta replaces at least one of the 10 strategies 3-4 times per year.
Our Approach to Weighting Strategies
Next, we have to determine how to weight our 10 strategies.
Equal-weighting would be one approach, but it fails to consider fundamental differences between our strategies. A well-diversified strategy like Adaptive Asset Allocation from ReSolve is a very different animal than a highly concentrated strategy like Dual Momentum from Gary Antonacci. Not better or worse mind you, just different.
We don’t want to consider historical returns in our weighting. We’re only designing a portfolio to succeed in the following month, and historical returns say nothing about expected returns over such a short period of time.
Our weighting is instead based on each strategy’s historical volatility and correlation, characteristics that tend to be more stable. We run our 10 strategies through two portfolio optimization routines: Equal Risk Contribution and David Varadi’s Minimum Correlation, and average the results. The former focuses on overweighting less volatile strategies (and vice-versa), while the latter on overweighting strategies less positively correlated to the others (and vice-versa). Why mix the two approaches? We see equal value in both, and want to capture characteristics of both. It’s important to note that strategies with a larger weighting are not “better” than those with a lower weighting. Strategies are weighted by how they “fit” with the other strategies selected.
Despite the fact that we’ve tried to balance the strategies selected, in any given month, the collective assets held by those strategies may be highly concentrated. That’s okay. If the collective opinion of these strategies is that the portfolio should be heavily weighted towards asset X at this moment in time, we want to capture that, as that’s the whole point of TAA.
Our Approach to Trading Meta Strategy
Now that we’ve selected our 10 strategies and how to weight them, we need to translate that into an asset allocation.
On the last trading day of each month, we take each of the 10 strategies selected for that month, and weight their asset allocations by the strategy’s allocation. Consider a simple two strategy example:
- 40% of the portfolio is allocated to Strategy A, and Strategy A is 50% long SPY and 50% long GLD.
- 60% of the portfolio is allocated to Strategy B, and Strategy B is 50% long SPY and 50% in cash.
- Meta Strategy will then be 50% long SPY, 20% long GLD and 30% in cash.
All of this is occurring in near real-time. Throughout the last trading day of the month, we’re updating the expected combined allocation at that day’s close.
We’ve added one small bit of additional sophistication:
Because each strategy is only contributing a small portion of the overall portfolio’s allocation, it’s common for a particular asset to have an impractically small allocation relative to the whole. So for any asset with less than a 3% allocation in the combined portfolio, we “roll that allocation up” into a higher level asset.
For example, exposure to emerging markets (EEM) would roll up to developed international markets (EFA), and exposure to the Nasdaq 100 (QQQ) would roll up to the S&P 500 (SPY). Again, this is only for combined allocations smaller than 3%. This approach has had essentially no impact on historical performance, but makes trading Meta Strategy much more practical for retail investors.
The aggregated nature of Meta means that it will never be the best performing TAA strategy in any given month or year, and it will never be the worst. That doesn’t mean that we’re trying to achieve TAA’s “average” performance. There is clearly some opportunity to separate out what could broadly be described as good strategies from less good strategies.
Our goal is to see Meta be in the top half of strategies in any given month or year. If it can do that consistently, then it will be a top performing strategy over the long-term, while removing the non-systemic risk of committing to any single strategy or approach.
We’ve leveraged our unique data set of published TAA models, coupled with unique analytical perspectives like “alternate trading days”, to create a unique product that we think really distills the best of tactical asset allocation.
We invite you to become a member for about a $1 a day, or take our platform for a test drive with a free limited membership. Put the industry’s best tactical asset allocation strategies to the test, combine them into your own custom portfolios, and then track them in near real-time. Have questions? Learn more about what we do, check out our FAQs or contact us.