The key takeaway: The Portfolio Optimizer is effective at selecting optimal strategy combinations, even when “walked-forward” (i.e. when limited to data it would have had at that moment in time).
First, a bit of background knowledge you’ll need to understand this analysis…
Background Knowledge: Model Portfolios and the Portfolio Optimizer
We track 90+ asset allocation strategies. Members can combine those strategies into what we call “Model Portfolios”. A Model Portfolio is like any other portfolio, but rather than combing assets, we’re combining strategies.
To help members determine the best way to combine strategies, we provide the Portfolio Optimizer. Members select an optimization objective like Max Sharpe or Min Variance, and we provide strategy combinations that have been optimal historically.
These optimizations are created with the benefit of hindsight. In other words, they assume that we had always traded the combination of strategies that were optimal knowing everything we know today.
Of course, that’s an unrealistic assumption.
From Hindsight to Walk-Forward:
How well would these optimizations have performed if they had instead been created in real-time? In other words, if they had been “walked forward” through history? If these portfolios would not have outperformed when created without the benefit of hindsight, then there’s no reason to trust today’s optimizations created with the benefit of hindsight.
We’re launching a new feature in April to fill in this missing piece. We’ll be adding 9 walked-forward optimized portfolios covering the major optimization objectives (Max Sharpe, Min Variance, etc.)
Each assumes that at the end of each year since 1972, we found the optimal combination of strategies based on data that would have been available at that moment in time. We then traded that portfolio from the end of January until the end of the following January. Repeat annually.
This is very resource intensive, so we don’t cover every possible iteration of the Portfolio Optimizer. We cover the “generic” (10 strategy) optimizations with no additional options selected – with a couple of exceptions. The 9 walk-forward optimizations are:
- Max Sharpe
- Max Sharpe w/ Low Exposure to Rising Interest Rates
- Max Sharpe w/ High Tax Efficiency
- Max Sortino
- Target Return = S&P 500
- Target Risk = 60/40 Benchmark
- Min Variance
- Max Diversification
- Min Correlation
Geek Note: Max UPI is not included in this analysis (yet). Max UPI is an especially resource-intensive objective because it relies on the “sequence” of returns. We’ve punted Max UPI to a future upgrade.
In the remainder of this article, we compare these walked-forward results to the top portfolio created by the Portfolio Optimizer today (during overlapping years).
Two principles:
Two principles to bear in mind when analyzing these results:
- Hindsight results will almost always be better than walk-forward results. That’s true for all historical prediction analysis. Predicting the past based on everything we know today usually outperforms predicting the past with only the data available at that time. The real question is how big that discrepancy is.
- Volatility and correlation are much easier to predict than return. For example, one would expect to see less discrepancy in walk-forward Min Variance results which only consider volatility and correlation, than Max Sharpe results which also consider return.
Walk-Forward vs Hindsight Results: Annual Return
For brevity, WF = Walk-Forward, HS = Hindsight
We start by comparing WF vs HS annual returns. This is one of the least informative of the statistics. None of the portfolios are optimized for “return”. At the most, they’re optimized for return relative to some measure of risk. So, whether WF results out/underperform HS doesn’t mean much.
| Walk-Forward vs Hindsight Portfolio Optimization: Annual Return 1973 to 02/2025 |
|||
|---|---|---|---|
| Optimization Objective | Walk Forward |
Hindsight | Diff (WF – HS) |
| Max Sharpe | 13.2% | 13.7% | -0.5% |
| Max Sharpe w/ Low Exposure to Rates | 13.3% | 14.2% | -0.8% |
| Max Sharpe w/ High Tax Efficiency | 10.8% | 10.5% | +0.3% |
| Max Sortino | 13.2% | 13.4% | -0.2% |
| Target Return = S&P 500 | 9.0% | 11.0% | -1.9% |
| Target Risk = 60/40 Benchmark | 15.1% | 16.2% | -1.1% |
| Max Diversification | 10.3% | 10.0% | +0.3% |
| Min Correlation | 10.8% | 11.0% | -0.3% |
| Min Variance | 8.5% | 8.1% | +0.4% |
| Average Optimization | 11.6% | 12.0% | -0.4% |
| 60/40 Benchmark | 9.3% | 9.3% | – |
Observations:
- WF and HS returns for objectives that don’t consider return – Max Diversification, Min Correlation, Min Variance – were very similar (WF was on average a bit higher).
-
The difference between WF and HS returns for objectives that consider return relative to risk tended to be greater, but not significantly so.The biggest exception is Target Return. This will be a common theme throughout this analysis. Target Return is the most “return oriented” of the optimization objectives. Overall, it failed in its stated mission of matching the return of the S&P 500, demonstrating how difficult it is to precisely target a specific % return.
Walk-Forward vs Hindsight Results: Sharpe Ratio, Sortino Ratio and Ulcer Performance Index
These are all measures of risk-adjusted return. A quick summary of each (note: “excess return” means return minus a risk-free rate):
- Sharpe Ratio = Excess return relative to volatility
- Sortino Ratio = Excess return relative to downside volatility
- Ulcer Perf. Index = Excess return relative to the length and depth of drawdowns
| Walk-Forward vs Hindsight Portfolio Optimization: Sharpe Ratio 1973 to 02/2025 |
|||
|---|---|---|---|
| Optimization Objective | Walk Forward |
Hindsight | Diff (WF – HS) |
| Max Sharpe | 1.19 | 1.34 | -0.16 |
| Max Sharpe w/ Low Exposure to Rates | 1.23 | 1.33 | -0.10 |
| Max Sharpe w/ High Tax Efficiency | 0.90 | 1.01 | -0.11 |
| Max Sortino | 1.22 | 1.33 | -0.11 |
| Target Return = S&P 500 | 0.96 | 1.24 | -0.28 |
| Target Risk = 60/40 Benchmark | 1.13 | 1.24 | -0.12 |
| Max Diversification | 1.09 | 1.16 | -0.07 |
| Min Correlation | 1.13 | 1.19 | -0.06 |
| Min Variance | 0.88 | 0.84 | +0.04 |
| Average Optimization | 1.08 | 1.19 | -0.11 |
| 60/40 Benchmark | 0.48 | 0.48 | – |
| Walk-Forward vs Hindsight Portfolio Optimization: Sortino Ratio 1973 to 02/2025 |
|||
|---|---|---|---|
| Optimization Objective | Walk Forward |
Hindsight | Diff (WF – HS) |
| Max Sharpe | 2.36 | 2.87 | -0.51 |
| Max Sharpe w/ Low Exposure to Rates | 2.52 | 2.86 | -0.33 |
| Max Sharpe w/ High Tax Efficiency | 1.61 | 1.94 | -0.34 |
| Max Sortino | 2.57 | 2.91 | -0.34 |
| Target Return = S&P 500 | 1.93 | 2.65 | -0.34 |
| Target Risk = 60/40 Benchmark | 2.23 | 2.42 | -0.19 |
| Max Diversification | 2.24 | 2.48 | -0.24 |
| Min Correlation | 2.32 | 2.47 | -0.14 |
| Min Variance | 1.69 | 1.58 | +0.11 |
| Average Optimization | 2.16 | 2.46 | -0.30 |
| 60/40 Benchmark | 0.80 | 0.80 | – |
| Walk-Forward vs Hindsight Portfolio Optimization: Ulcer Performance Index (UPI) 1973 to 02/2025 |
|||
|---|---|---|---|
| Optimization Objective | Walk Forward |
Hindsight | Diff (WF – HS) |
| Max Sharpe | 4.58 | 6.39 | -1.81 |
| Max Sharpe w/ Low Exposure to Rates | 5.19 | 6.00 | -0.81 |
| Max Sharpe w/ High Tax Efficiency | 2.68 | 3.24 | -0.56 |
| Max Sortino | 5.13 | 6.47 | -1.35 |
| Target Return = S&P 500 | 3.41 | 6.03 | -2.62 |
| Target Risk = 60/40 Benchmark | 3.62 | 5.06 | -1.44 |
| Max Diversification | 5.20 | 6.24 | -1.04 |
| Min Correlation | 4.82 | 5.51 | -0.69 |
| Min Variance | 3.28 | 3.01 | +0.26 |
| Average Optimization | 4.21 | 5.33 | -1.12 |
| 60/40 Benchmark | 0.78 | 0.78 | – |
Results were generally similar to what we saw with annual returns. WF and HS results were more similar for “non-return” objectives, and the difference between WF and HS results was particular large for Target Return.
One additional observation:
There was a much larger difference between WF and HS results for Max Sharpe than Max Sharpe w/ Low Interest Rate Exposure. That disparity is largely a result of the 2022 market downturn, when there was a concurrent loss in both risk assets and bonds not seen in nearly 100 years (read more). Bonds tend to offset big losses in risk assets, but that wasn’t the case that year.
For Max Sharpe w/ Low Rate Exposure, that’s not as big of a deal, because it tends to pick strategies that are less likely to hold long duration bonds anyways.
But the generic Max Sharpe, in real-time, picked strategies with too much exposure to long duration bonds, because that’s what worked historically. It’s only in hindsight that it sees that was a bad decision, meaning in hindsight, it was less likely to pick these strategies.
That’s an especially clear example of the importance of walk-forward analysis, and why it more accurately captures true performance.
Walk-Forward vs Hindsight Results: Annual Volatility
This is the least exciting stat we’re looking at. Having said that, all of these optimization objectives in some way consider volatility, so it’s important that they successfully manage it on a walk-forward basis. Note: Unlike other stats discussed, a lower value is preferred.
| Walk-Forward vs Hindsight Portfolio Optimization: Annual Volatility 1973 to 02/2025 |
|||
|---|---|---|---|
| Optimization Objective | Walk Forward |
Hindsight | Diff (WF – HS) |
| Max Sharpe | 7.2% | 6.8% | +0.4% |
| Max Sharpe w/ Low Exposure to Rates | 7.2% | 7.3% | -0.1% |
| Max Sharpe w/ High Tax Efficiency | 6.9% | 5.9% | +1.0% |
| Max Sortino | 7.1% | 6.7% | +0.4% |
| Target Return = S&P 500 | 4.7% | 5.2% | -0.5% |
| Target Risk = 60/40 Benchmark | 9.4% | 9.3% | +0.1% |
| Max Diversification | 5.3% | 4.7% | +0.6% |
| Min Correlation | 5.5% | 5.4% | +0.1% |
| Min Variance | 4.5% | 4.2% | +0.3% |
| Average Optimization | 6.4% | 6.2% | +0.3% |
| 60/40 Benchmark | 10.0% | 10.0% | – |
No significant observations here.
Annual volatility is most relevant to the non-return objectives (Max Diversification, Min Correlation and Min Variance) and in all cases, WF and HS were reasonably close.
The biggest discrepancy was Max Sharpe w/ Tax Efficiency. We don’t have a clear explanation as to why. This objective is pulling from the smallest set of strategies (strategies w/ long-term tax efficiency >= 75%), so perhaps there’s a little less consistency in the composition of portfolios over the years?
Walk-Forward vs Hindsight Results: Max Drawdown
Max drawdown is like annual return in that none of these objectives is directly optimizing for this stat, but it’s still interesting to see the results.
| Walk-Forward vs Hindsight Portfolio Optimization: Max Drawdown (EOM) 1973 to 02/2025 |
|||
|---|---|---|---|
| Optimization Objective | Walk Forward |
Hindsight | Diff (WF – HS) |
| Max Sharpe | -10.5% | -7.1% | -3.4% |
| Max Sharpe w/ Low Exposure to Rates | -7.8% | -7.7% | -0.1% |
| Max Sharpe w/ High Tax Efficiency | -11.6% | -9.5% | -2.1% |
| Max Sortino | -7.6% | -6.3% | -1.2% |
| Target Return = S&P 500 | -5.1% | -5.7% | +0.6% |
| Target Risk = 60/40 Benchmark | -11.5% | -10.6% | -0.9% |
| Max Diversification | -6.8% | -4.9% | -1.9% |
| Min Correlation | -5.6% | -6.7% | +1.1% |
| Min Variance | -4.8% | -4.7% | -0.0% |
| Average Optimization | -7.9% | -7.0% | -0.9% |
| 60/40 Benchmark | -29.5% | -29.5% | – |
These WF vs HS results were surprisingly similar. Max Drawdown can be a very volatile metric, because it’s measuring just one single worst moment in time.
The biggest difference between WF and HS was in Max Sharpe for the reasons we discussed previously re: the 2022 market downturn.
Walk-Forward vs Hindsight Results: Wrap Up
Individual optimization objectives differ, but broadly speaking, when comparing WF to HS results since 1973, annual return has fallen by ~0.4%, annual volatility risen by 0.3%, Sharpe, Sortino and UPI fallen by 0.1, 0.3 and 1.1 respectively, and Max Drawdown worsened by 0.9%.
The least impacted objectives are those that are non-return-oriented: Max Diversification, Min Correlation and Min Variance. The most impacted objective is Target Return = S&P 500. Based on these results, we would be hesitant to employ this particular optimization.
Again, it should not be surprising that there is some deterioration in performance. Hindsight results will generally outperform walk-forward results. The more important takeaway is the fact that these more adversarial walk-forward results are still so strong. That’s a major feather in the cap of using these type of optimization techniques to build Model Portfolios.
Next steps:
As mentioned, we’ll be adding these 9 walked-forward optimized portfolios to the platform in April. For members who don’t want the complication of handcrafting their own Model Portfolio, these walked-forward optimizations will serve as a simple all-in-one solution that can be followed directly like any other strategy.
For members willing to take on more responsibility, the existing Portfolio Optimizer is still the best resource for creating optimized portfolios because members have more control over the results.
New here?
We invite you to become a member for about a $1 a day, or take our platform for a test drive with a free membership. Put the industry’s best Tactical Asset Allocation strategies to the test, combine them into your own custom portfolio, and follow them in real-time. Learn more about what we do.
