top of page

Analysing risk: horses for courses

Updated: Jun 8, 2023

Understanding the challenge

We’ll start with an irrefutable fact: fund return volatility is required to generate decent performance.

Investment strategies with decent capacity which generate exceptionally high information ratios don’t exist or are extraordinarily rare.

The challenge then is not so much to minimise risk as to take on the most appropriate type of risk, given the investment strategy.

Consider two diametrically opposed investment strategies: fundamental long/short and risk-arb.

A fundamental long/short manager should be cognisant of factor bets (eg factor exposures based on Barra’s risk model). The portfolio management team should focus on stock selection rather than taking factor bets. However, there needs to some form of “risk”. For a long/short manager, it’s typically stock concentration risk. The portfolio management team may argue that this risk is manageable given their in-depth analysis. They may also have stock specific risk controls in place, such as a stop-loss limit. Regardless, taking concentrated stocks bets is risky. Anyone who doubts this should review the events leading to the recent demise of Melvin Capital.

A risk-arb manager should also have minimal factor bets, outside of momentum. This strategy also necessitates taking numerous small positions – hence there shouldn’t be any stock concentration risk. Once again, however, there needs to be some form of “risk”. For a stat-arb manager, it’s mainly leverage risk. This is more of a “tail risk” in that when markets are well behaved it’s not problematic. However, during market dislocations and extreme liquidity events, leverage risk comes to the fore. Witness the demise of LTCM in 1998 and

the downfall of numerous quant funds during the Great Quant Unwind in August 2007.

What risks are appropriate for our investment strategy?

First, we’ll rule out what risks are inappropriate:

  • Stock concentration risk: We play a “probability” rather than a “perfection” game. Alpha factors with huge predictive power don’t exist. Their primary function is to tilt the odds in your favour.

  • Leverage risk: Our view on leverage differs from many other quant fund managers. It’s based on many years of experience and what we witnessed during the Great Quant Unwind. Excessive leverage can result in forced unwinds during the most inopportune times, undoing many years of alpha generation. We will never put ourselves in this position.

  • Market beta: No surprisingly, this is the biggest risk factor. Further, market timing isn’t suited to quant modelling due to the lack of breadth. Based on the Fundamental Law of Active Management, you would need a model with huge predictive power. Backtesting quant models with limited degrees of freedom is also prone to data mining.

Next, we’ll define quant factors and make a distinction between alpha factors and risk factors. A quant factor is any measurable stock attribute which explains (ie is correlated) with cross-sectional returns. Alpha factors tend to be positively correlated with stock returns and, importantly, there is a reason why they exhibit predictive power. Risk factors are correlated with stock returns, but sometimes the correlation is positive and sometimes it’s negative. There’s also no reason why they should exhibit consistently positive predictive power.

Complicating matters further, some factors exhibit both alpha and risk characteristics – eg. value and momentum – in that over the long term they are positively correlated with stocks returns but are prone to periods of underperformance and, in the case of momentum factors, the performance reversal can be very swift and severe.

So, what do we do?

In broad terms, we take informed alpha factor bets while minimising risk factor bets.

After accounting for market beta, the key risk factors are country, sector, and size. There are myriad factors which explain stock returns univariately. However, using a multi-variate framework, after accounting for country, sector and size, other factors don’t exhibit statistically significant predictive power.

For country, we use “country of risk” rather than “country of listing”. This considers where companies source most of their revenue and where key employees are located.

For sector, we monitor GICS1, GICS2 and GICS3, both within markets and across the portfolio.

For size, we divvy the long and short portfolios into different groupings based on market cap and liquidity.

Importantly, we will take risk tilts (eg sector tilts). If we didn’t do this, we would have to take on an unacceptable level of leverage risk.

To help explain how we view our risk tilts, we’ll use a rubber band analogy. We allow the rubber band to stretch, but not to a breaking point. Hence, we have hard risk limits (eg sector limits). Also, our position sizing methodology makes it more difficult for risk tilts to get bigger. The rubber band can stretch but it requires more force to do so and there’s a limit to how far it can go.

There’s no limit on the size of alpha factor bets. However, we run a multi-factor process and many of the factors are inversely correlated. Hence, no particularly alpha factor bet will drive portfolio performance. In Z score space, our long portfolio nearly always has a positive exposure to every key alpha factor and composite, and our short portfolio nearly always has a negative exposure to these alpha factors and composites.

Our alpha factors include value and momentum factors. They’re more sophisticated than the simple value and momentum proxies used by Barra, but they’re correlated with these factors.

We get our exposure to these factors in a sophisticated way via our stock selection and rebalancing process. Our preferred approach to combining factors is non-linear. We run stocks screens that include factors that work well together. For example, we combine value with certainty and profitability factors to avoid value traps.

In essence, the most appropriate type of risk for our strategy is alpha factor bets. These bets are based on sophisticated factors, and we get exposure to these factors using a sophisticated investment process. They are expressed via a large number of positions to

mitigate stock concentration risk.

Analysing our risk profile using a simplistic factor-based model that includes alpha proxies that we endeavour to exploit doesn’t work for our investment process. It’s the appropriate risk lens for long-short fundamental strategies and many other investment strategies, but not for us.

90 views0 comments

Recent Posts

See All

Investor sentiment & AI as a risk factor

“In the short run, the market is a voting machine but in the long run, it is a weighing machine.” This is a quote from Warren Buffett in 1987 who paraphrased a passage from Graham and Dodd’s seminal b


Функцію коментування вимкнено.
bottom of page