Risk Parity

I was going through some old posts and realized something surprising: I’ve never written a dedicated piece on risk parity. I could’ve sworn I covered it at some point—maybe under the All Weather umbrella—but apparently not. Chalk it up to the hazards of writing too much (or having a bad memory).

Anyway, I came across a paper on BankerOnWheels recently. It’s not exactly Nobel-worthy research, but it did raise a few interesting points that are worth unpacking. So let’s use it as a jumping-off point to talk about risk parity—at least, some things about it.

And while we are on the topic of poorly written papers, a quick public service announcement: if you’re burying all your charts and tables in the appendix, there is a special place in hell for lads like you.

“Merton (1980) and Chopra and Ziemba (1993) demonstrate that expected returns are significantly more difficult to estimate accurately than expected risks. Thus an asset allocation approach which ignores expected returns might benefit. This is, of course, what risk parity does.“

Given how short our investing lives really are, it’s probably easier to bet on stocks dropping 30% at least once over the next decade than to predict their terminal value with any real accuracy. That’s part of why I find the core idea behind risk parity so compelling. Returns come from taking risk—that’s the deal. But if I can diversify those risks intelligently, I give myself a better shot at staying invested through the inevitable volatility and actually capturing those returns over time. The key, of course, is choosing the right kinds of risk.

The paper quotes these four reasons as to why investors might want a risk parity portfolio:

1. The risk parity portfolio is something that helps institutions become more diversified.
2. The risk parity portfolio is also sometimes a mean-variance efficient portfolio.
3. The risk parity portfolio exploits the aversion investors maintain toward leverage.
4. The risk parity portfolio is more robust to estimation risk.

We can only identify the mean-variance optimal portfolio in hindsight. It’s a moving target we’ll never hit in real time. Still, there’s value in aiming for it—as long as we’re honest about the fact that we’ll never get there perfectly. It’s a bit like investing by elimination: the via negativa approach. Avoid the obvious mistakes, steer clear of what doesn’t work, and you’ll improve your odds by default.

The greatest advantages of the risk parity approach are 1) and 3), for yourstruly: it pushes to understand what being really diversified means (not US stocks and International stocks) and it demystifies the use of leverage. Point 4) is a mixed bag, in the sense that there is no ‘retail’ tool to backtest estimated inputs (more on this later).

The paper correctly highlights that there are many varieties of risk parity because risk can be defined in different ways:

• historical volatility
• expected volatility
• other measures of risks
• exposure to risk factors instead of asset classes

Historical volatility is what we have to deal with as retail investors, but it is important to stress that it is a far from perfect proxy of risk. For example, if we assume that stock returns are not i.i.d., then allocating more to stocks than what their volatility indicates would make sense (if we are looking at a long enough horizon)

Expected-returns-adjusted risk-parity portfolio

Just a quick note on something that struck me as odd. As I’ve written before here—and the paper itself even acknowledges—estimating correlations and volatility tends to be a lot more reliable than forecasting returns. That’s one of the reasons risk parity models typically focus on risk metrics rather than expected returns.

But in this case, the authors go in the opposite direction. They build a model that tries to forecast returns, and here’s the strange part: they set each asset’s expected return equal to its trailing 10-year average. That’s… a choice. Especially for bonds.

With bonds, today’s yield is generally the best predictor of future returns. They don’t mean-revert the way stocks do, so relying on a backward-looking 10-year average seems like a pretty shaky foundation.

Testfol.io

I’ve never had the chance to fully test a risk parity strategy myself, but I noticed that testfol.io recently added a new Optimiser feature. It’s a pretty simple version—it calculates portfolio weights based on input volatility and correlations, then keeps those weights constant throughout the backtest.

That’s fine as a starting point, but it’s worth noting that real-world risk parity strategies are dynamic. They adjust over time as volatility and correlations shift—metrics that can be estimated from recent data, like trailing 12-month volatility (in particular, when one asset’s volatility goes up, the model cuts exposure to that asset—that’s the way I understood it). More advanced models go even further by trying to forecast future volatility, correlations, and even expected returns (though that last part is notoriously tricky).

The ‘classic’ risk-parity portfolio has four ingredients: stocks, bonds, commodities and currencies. There is no ETF that we can use to proxy the currency ingredient, so our test is a bit muted.

Here is a test with gold and commodities:

and here one without gold:

Here is an experiment with my favourite mix of assets: stocks, bonds, gold and trend

and here is the Sharpe-optimal non-risk-adjusted portfolio:

The last example is the best possible portfolio with 20/20 hindsight. If I understood the tool correctly, the risk parity option builds the portfolio just by looking at assets’ volatility and correlations: it is a foundation that brings us pretty close to the optimal weightings.

Let’s do a different experiment. Here are the weights for the optimal risk parity portfolio:

I want to compare it with a really naive allocation and the 60/40:

The Naive Portfolio performance is not that bad! At the end of the day, the assets you choose to include in your portfolio will be the biggest driver of your risk-adjusted returns. Everything else—optimization, weighting, rebalancing—comes second.

The paper’s conclusions – pretty shallow?

Given how those two asset classes performed, it’s not exactly shocking that allocating more to the better-performing one—especially on a risk-adjusted basis—led to stronger returns.

But that’s kind of the point. Back in 1951, if you had no idea what the future Sharpe ratios of stocks and bonds would be, would 60/40 have seemed obvious? Probably not. That mix only became iconic after it delivered great results. It’s the same reason people started declaring it dead after 2022—it’s all hindsight.

What this paper shows isn’t that naive risk parity doesn’t work, but that even simple variations in weighting—based on basic formulas—can deliver strong outcomes. And from my own (even more naive) testing, it seems you might not need to tweak weights at all—just focus on actual diversification.

Of course, the valid pushback here is selection bias. Am I cherry-picking assets that happened to work well? Maybe. Would I have known a decade ago that long-only commodities wouldn’t perform? Probably not. Is trend following doing well in backtests just because the models that failed quietly disappeared? Also possible.

But let’s be honest—60/40, the S&P 500, the Nasdaq… these are all cherry-picked too. And I wouldn’t bet anything I value on the Nasdaq repeating its last 20 years over the next 20.

The Real World

At first glance, the performance of the actual risk parity funds highlighted in the paper felt a bit underwhelming—especially considering these are active strategies that can adjust weights and use forward-looking inputs. You’d expect a little more.

But then again, we’re looking at net-of-fee results across a mix of serious institutional products and a few that seem more like marketing exercises. Aggregating all that data can only tell you so much… or maybe I’m still giving the industry more credit than it deserves.

The RPAR ETF, for example, is built based only on long-term historical volatility and has exposure to the following categories:

• global stocks
• Treasuries
• TIPS
• commodity producers and gold

You can notice relevant differences from other ‘risk parity concepts’:

• Instead of commodities, they use commodity-related stocks
• TIPS were not part of the original concept (but they might useful! but they haven’t so far!)
• Each asset is not scaled down when volatility explodes
• There is no forward-looking / expected metric

One day, someone will do a paper comparing the risk parity principles with available retail products and I am sure they are going to include this…even if they shouldn’t.

And talking about real-life portfolios, StateStreet recently launched a risk parity ETF in collab with Bridgewater. For what it is worth, here is how it performed so far, compared to our other portfolios:

What I am reading now:

Follow me on Bluesky @nprotasoni.bsky.social