Abstract: In the past several weeks, risk parity funds such as Bridgewater’s All Weather have attracted substantial criticism. The attacks have clustered around two primary notions: firstly, that risk parity is a failure because it did not protect investors during August’s drawdowns and secondly, that risk parity may have caused or at least amplified the market’s sudden losses. In this post, we’ll evaluate the legitimacy of these two claims, as well as provide a deeper understanding of the motivations and behaviors of risk parity funds.
Defining Risk Parity
Most investor portfolios are heavily weighted towards equity risk. A typical 60/40 portfolio, for example, will usually derive 80-90% of its risk from equities since stocks are much more volatile that bonds.
Risk parity is an asset allocation methodology that strives to overcome this problem by creating portfolios that distribute risk roughly equally between assets. There are several ways to do this, but the most common approach is to weight assets by the inverse of their volatility. Doing so increases weights on low volatility assets and decreases weights on high volatility assets so that each asset’s levered or de-levered position has an identical expected volatility.
Most risk parity funds then lever this volatility weighted portfolio to target a specific level of risk. In Bridgewater’s case, this is a 12% annualized standard deviation. The result is a portfolio that has similar total risk to a 60/40 portfolio, but a much more equitable internal distribution of risks.
Risk Parity is a Simple, but Imperfect Solution to a Complicated Problem
The idea behind risk parity is a good one. It results in meaningfully more diversified portfolios. Risk parity is not a perfect strategy, however. It has plenty of faults, including:
Risk Parity Does not Account for Expected Returns
Risk parity makes the implicit assumption that reward-to-risk is roughly equal across assets and asset classes. This may be true, but it’s most likely not, especially if we’re going to define risk as volatility. The Capital Asset Pricing Model argues that reward-to-beta should be equal, but beta is not the same thing as volatility. And CAPM is an empirical failure, anyway, which is why people are so often developing new factor models to explain asset behavior.
Risk Parity Ignores Correlations
This is a major issue. We can throw a bunch of assets in a blender and equal volatility weight them, but if many of them are highly correlated we can wind up with a risk profile that is not at all balanced. As an example, consider a risk parity portfolio constructed using US equities, foreign developed equities, emerging market equities, corporate credit spreads, emerging market bonds, and US Treasuries. All of the non-Treasury assets share a similar risk profile and offer only marginal diversification benefits. An unconstrained risk parity portfolio employing them would wind up with about 5/6 equity risk and 1/6 fixed income risk, exactly the situation risk parity is trying to avoid.
Most risk parity strategies attempt to overcome this problem by segmenting assets into buckets and enforcing parity within and among those buckets.
All Weather segments first by market environment, then by asset, so that the resulting portfolio is more or less probability weighted to handle most major financial conditions. That’s certainly not a bad way to do it, but it does introduce more subjectivity into the process.
A better way to achieve risk parity’s goals would be to use equal risk contribution weights instead of inverse volatility weights, but doing so would require us to estimate correlations as well, which introduces another layer of subjectivity. Furthermore, correlations, like volatilities, are non-constant.
Risk Parity Assumes that we can Capture Risk in a Single Measure, and that Volatility is that Measure
Risk parity ignores higher moments like skew and kurtosis. It ignores tail risk. It ignores duration and convexity, prepayment risk, default risk, etc. That’s a lot to ignore. When you size your positions based solely on volatility you can easily wind up with a portfolio embedded with massive hidden iceberg risks. Consider, for example, if you were to include an out-of-the-money put selling strategy in a risk parity portfolio. The strategy’s low volatility would result in it having a very large weight, and a decent sized drawdown would leave you penniless and ruined.
Risk Parity Relies on Subjective Estimates of Volatility
There is no universal standard for estimating volatility. Do you use realized volatility? Over what period? Do you incorporate rolling windows, expanding windows or exponential decay? How about implied volatilities? With what option expiration? Or should you use a GARCH model? How frequently do you update your estimates and how much do you weight the current market environment versus the long-term mean? There is no single correct answer. And even if there were, forecasts of future volatility would still be error prone.
Investigating Claim #1: Is Risk Parity a Failure?
Thus far we’ve established that risk parity is fallible, but does that fact, combined with August’s losses, mean that risk parity is a failure?
Absolutely not. All Weather’s name may give the impression that it is designed to profit over any time period in any market environment, but that’s not the case. All Weather seeks to balance risk through four likely market environments over the long-term. It does so through diversification, not hedging. It is not an absolute return strategy, and it is certainly not a tail risk solution.
Investors need only to look at All Weather’s historical record to confirm this. It lost 31.6% in September and October of 2008 and has had drawdowns in excess of 10% three times since its inception in 1996. It is, by no means, a strategy intended to avoid all losses.
All Weather’s bets are made far in advance. It makes no explicit short-term forecasts. Rather, it strives to keep risk weights constant over time regardless of market conditions. There are no tactical adjustments other than resizing positions based upon updated volatility assumptions.
When risk assets lose value and bonds fail to compensate, All Weather is going to lose money. When markets blow up, correlations go to 1, and bonds do nothing, All Weather is going to lose a lot of money.
This is precisely what happened in late August. Investors shared two very large concerns: firstly, that global growth would slow down as a result of China’s implosion and secondly, that the Fed would soon be raising interest rates. That combination is bad for everything: stocks, corporate credit, commodities, bonds, you name it. Add in the fact that China was dumping Treasuries to support the Yuan and it’s not hard to see why bonds failed to rally as risk assets fell off a cliff.
Were All Weather’s August Returns Reasonable?
Since we know All Weather’s approximate risk allocation we can, as a sanity check, estimate what its return should have been given market returns.
Our estimate of -3.75% is a bit better than Bridgewater’s actual performance, but it’s nevertheless in the ballpark. The final levered return projection varies significantly based on our volatility estimates. We used a variety of estimation methods and produced returns as high as -2% and as low as -6%. Bridgewater’s return is well within that range, so you can’t really argue that they’ve executed the risk parity strategy poorly.
If you’re going to criticize Bridgewater for anything, it should probably be for failing to include a declining growth/rising interest rate scenario in its environment matrix, but to be honest there’s not a lot out there asset-wise that would have protected investors in that case. Long variance, short emerging currencies?
Any competent investor should have been able to see in advance that this economic scenario was one of risk parity’s greatest weaknesses. By investing in the strategy investors were implicitly accepting this risk. The fact that they got burned by it is not Bridgewater’s fault. It is not risk parity’s fault. It is entirely their own. They should accept that. Most likely do.
Investigating Claim #2: Did Risk Parity Exacerbate the Market’s August Decline?
Much has been said about the fact that volatility targeting gives risk parity a momentum-like component. When an asset’s price falls, its volatility often increases. The higher volatility means that a risk parity portfolio needs less of the asset to meet its constant volatility target. This means that the portfolio needs to sell the asset, which may further depress the asset’s price.
Marko Kolanovic of JPMorgan seized upon this fact while arguing that risk parity funds and CTAs contributed significantly to the pre-market flash crash on Monday, August 24th. Leon Cooperman of Omega Advisors went even further, attempting to shift the blame for his funds’ 9%-11% losses last month.
But how legitimate are these claims?
Kolanovic is correct that risk parity funds will need to sell assets if their volatility expectations increase, all else equal. The question then is, how much did volatility expectations increase? Kolanovic and others seem to be assuming that risk parity funds would adjust their volatility assumptions roughly in line with a short-term measure like the VIX, which jumped from 13.79 on August 18th to 40.74 on the 24th. The VIX fell back to 26.10 three days later, only a bit higher than where it is right now. If the VIX doubled or tripled that must mean that risk parity funds cut their exposure by half or two-thirds, right?
Volatility Tends to Mean Revert and Risk Parity Funds Manage to a Longer-Term Target
Volatility has an interesting property: it is mean reverting. When it is depressed, it has a tendency to rise. Conversely, when it spikes, it has a tendency to fall. Short-term volatility measures do a poor job of predicting long-term volatilities, something that Bridgewater and other risk parity practitioners are well aware of.
Risk-parity funds are not like levered ETFs, which recalibrate their risk daily. They manage to a target level of volatility over the long-term, not to the day. While they probably did adjust their forward-looking volatility estimate upwards, they almost certainly did not do so at the same magnitude as the increase in the VIX.
To get an idea of how much a longer-term perspective may have dampened volatility increase expectations, we can look to the VIX futures market. In the following table we look at price increases along the VIX futures term structure and the position reduction a risk parity fund would have had to undertake to retain a constant volatility position if it were using that point on the curve as its estimate of future volatility.
It is very evident that having a longer-term view dampens volatility adjustments quite significantly. A risk parity fund basing its volatility assumption on the VIX and fully adjusting daily would have had to reduce its equity exposure by more than 66% on August 24th. A risk parity fund basing its assumption instead on the 6-month February VIX futures would have required a reduction of only 13.4%, roughly one-fifth as much.
If Bridgewater or other risk parity funds were basing their volatility assumptions on implied volatility out longer than a few months, the impact on markets would be a small fraction of JPMorgan’s projections.
Risk Parity Funds do not Want to Adjust Volatility Expectations Too Quickly
Risk parity critics are correct when they assert that maintaining a constant level of volatility introduces momentum-like properties, however they fail to consider that momentum strategies typically only work at medium-term horizons, usually somewhere between one month and one year. There is actually an opposite effect at the daily and weekly level, one that is perhaps even more economically significant: mean reversion.
Assets experiencing abnormal gains and losses over short periods tend to reverse those moves over subsequent days. A quantitatively-oriented fund such as Bridgewater would be well aware of this fact.
If Bridgewater were to aggressively recalibrate their volatility assumptions based on short-term measures such as the VIX, their momentum-like trading would actually be shorting this mean reversion effect, which over time would cause a material amount of portfolio bleed.
And that’s not even including the additional transaction costs that frequent rebalancing would incur.
Normal Risk Parity Rebalancing Produces an Anti-Momentum Effect
Absent any changes in volatility expectations, risk parity rebalancing actually results in a stabilizing effect, illustrated in the following tables.
Here we look at a two-asset risk parity portfolio comprised of stocks and bonds. We will target a 12% total portfolio volatility, much like All Weather does. We will assume that stocks have a long-term volatility of 16% and that bonds have a long-term volatility of 6%. Risk parity calls for an unlevered portfolio that is weighted 27% to stocks and 73% to bonds. To obtain our target of 12% total portfolio volatility we need to lever the unlevered portfolio 1.94 times, such that our final weights are 53% stocks and 141% bonds.
To make it simple, we will assume that bond and stock futures are both priced at $1 and that our portfolio has $1 in cash. To set up our target portfolio we then need to buy 0.53 stock futures and 1.41 bond futures.
When stocks fall, bonds increase and volatility remains unchanged, the portfolio becomes overweight bonds and underweight stocks. It therefore needs to sell bonds and buy stocks to reacquire its target exposure, therein buying the underperforming asset and selling the outperforming asset.
In our scenario, after a 5% loss in stocks and a 1% gain in bonds, the fund needs to increase its stock position from 0.53 futures to 0.55 futures, or approximately 4%. This isn’t a lot, of course, but it is meaningful.
We considered a case where volatility remained constant. If volatility assumptions were to instead increase, this required stock purchase would net against the sales required to maintain the portfolio’s target volatility, thereby reducing the size of the trade we need to execute externally in the market.
The Trades Required by Risk Parity Would have been Small Compared to Total Volume Traded
If, for a moment, we suppose that JPMorgan’s assumptions are correct, and that risk parity funds adjusted their volatility assumptions based on short-term volatility measures, that would leave them with $100 billion in equity index exposure to sell (JPMorgan’s numbers, not ours; Bank of America estimated $30-80 billion).
How would that pressure affect the market?
Firstly, risk parity funds tend to source their equity risk globally, so they would only need to sell about half of that $100 billion domestically, likely through S&P 500 futures. For a bit of perspective, we can look at recent futures trading volumes.
The following table shows billions of notional futures traded before and through the market’s August drawdown. SP represents the CME’s big S&P futures contract while ES represents the more liquid e-mini.
In the 20 trading days ending on August 18th, the ES contract averaged $158 billion dollars in notional volume each day. The SP contract added another $2 billion for a total of $160 billion. On Monday the 24th, volumes spiked substantially, with a total of $503 billion traded. For the full week starting on the 24th, the daily average was $340 billion.
Even if risk parity funds traded that entire $50 billion on the 24th, it would have represented only 10% of the total volume for that day and 14.6% of the excess volume above the prior period’s mean.
If risk parity funds instead spaced their trades out over a period of three weeks, as JPMorgan suggests, that number falls to around 1% of total volume per day, which is de minimis.
If we go further to assume that risk parity funds base their volatility estimates not on the VIX but on longer-dated implied volatilities, then that 1% declines even further. Previously we calculated that using 6-month VIX futures prices instead of the VIX would reduce trade sizes by four-fifths. That would leave us with 0.2% of futures volume per day.
And we haven’t even considered the numerous other vehicles risk parity funds could use to reduce their equity exposure, such as ETFs, individual stocks, index options, and OTC derivatives.
It is impossible to logically conclude that risk parity could alone account for the market’s fall. It’s also very difficult to argue that risk parity significantly exacerbated the move.
Risk Parity is a Small Portion of the Money Management Industry
According to JPMorgan, risk parity funds have approximately $500 billion in assets under management. BarclayHedge estimates that hedge funds collectively had $2.7 trillion in AUM at the end of June. The Investment Company Institute estimates that mutual funds had $16 trillion in AUM at the end of 2014 and that ETFs held anther $2.7 trillion.
Here’s what that looks like in a pie chart:
The hedge fund, mutual fund, and ETF industries, collectively, are about 42 times the size of risk parity. And that’s not even including the trillions more that are owned by institutional and retail investors outside of structured investment vehicles. Suggesting that risk parity could move the market sounds suspiciously like the tail wagging the dog.
I have never, in my life, met a competent trader that did not reduce risk in times of uncertainty. Are we really to believe that risk parity funds were the sole agents to react and that every other market participant stood idly on the sidelines? That’s simply ludicrous.
Active managers had significant reasons to reduce their risk when the market began trending downward. Those evaluated on a relative basis would seek to reduce tracking error relative to their benchmark. Those evaluated on an absolute basis would seek to preserve capital. There is no reason to believe that risk parity funds would ever act alone, and the $343 billion of extra futures volume on August 24th confirms that.
Risk parity is not without its flaws, but recent criticisms have little to support them. Investors may or may not agree with the philosophy behind risk parity, but they can’t reasonably argue that Bridgewater failed in its execution in August or that risk parity should shoulder the blame for the market’s woes.