Increasingly, investors are placing a lot of emphasis on idiosyncratic or stock specific risk. Depending on how this risk is calculated, it should be interpretted with caution.
Here’s our list of potential issues:
Comingling risk and alpha factors
Ignoring stock selection
Overlooking the benefits of combining factors using a robust multivariate framework
Ignoring the benefits of stock diversification
Overlooking the risks posed by high leverage
Not analyzing country risk properly.
The relevance of these issues varies for different investment methodologies. For a fund like ours, they are all relevant and hence worthy of detailed analysis.
Comingling risk and alpha factors
Factors explain cross-sectional stock returns. Based on univariate regressions, there are myriad such factors. However, using a stepwise regression approach, starting with the factors with the strongest predictive power, the list of statistically significant factors quickly dwindles.
I’ve never used Barra or Axioma, but I believe the factors they use are the same as the key ones we’ve identified based on our modelling. The list is intuitive and commonly referenced in academic literature:
Country
Sector
Size
Value
Momentum.
We now need to make an important distinction between risk and alpha factors. Risk factors explain cross-sectional returns, sometimes positive and sometimes negative. Alpha factors also do this but, on average, over time, they are positively correlated with stock returns. Importantly, there should also be a reason why these factors work, typically based on investors’ behavioral biases.
Returning to our list of factors, value and momentum satisfy our alpha criteria. Admittedly, the factors used in risk models, by both practitioners and academics, tend to be very simplistic, but they are still correlated with more sophisticated factors. Having a positive net exposure to these factors isn’t a cause for concern, especially if it’s achieved in a smart or sophisticated way. This brings us to the next section.
Ignoring stock selection
Maybe you’re thinking that anyone with a spreadsheet could select stocks based on simple value and momentum factors – and you’re right. So why should you pay hedge fund fees for this simple task?
A good fund manager will generate alpha by ignoring false positives – eg cheap stocks which are value traps – and more broadly by selecting stocks which are compatible with the mispricing opportunity that is being targeted (and weighting the stocks accordingly).
We will illustrate this point by focusing on a simple value factor: one year forward earnings yield. Let’s assume there are 5 value managers who are selecting long (or overweight) positions based on the top quartile of stocks ranked by this factor and short (or underweight) positions based on the bottom quartile.
Let’s also assume that the skill level of these managers varies significantly, from exhibiting a high level of skill to selecting stocks randomly. (We do this by assuming perfect foresight and then applying a “noise” factor with varying levels of intensity. The noise factor varies from 20 for the best manager which assumes very strong foresight, to 1,000 for the worst manager which equates to largely random stock picks. The portfolio is rebalanced monthly using each way transaction costs of 15bps. The universe is the point-in-time top 500 stocks in Japan).
The cumulative return (Chart 1) and information ratio (Chart 2) for the 5 managers are shown below.
Chart 1: VAMI of Different PMs Source: OQFM
Chart 2: Information Ratio for different skill levels Source: OQFM
Manager A does a great job of picking the best cheap socks and the worst expensive stocks. We also strive to do this using our sophisticated value models and stock selection process.
One thing each of the managers have in common is the long and short net exposure to one year forward earnings yield. To illustrate: the average Z score exposure to one year forward earnings yield for Manager A’s long portfolio is 1.34, while for Manager E it’s 1.33. Conversely, for the short portfolio the average Z score exposure for Manager A and Manager E are –1.41 and –1.40 respectively.
Given this, the idiosyncratic risk will be dampened to the same degree, despite the extreme difference in the skill level and overall attractiveness of the managers. This clear illustrates how idiosyncratic risk can fail to differentiate between good and bad quant managers.
Overlooking the benefits of combining factors using a robust multivariate framework
A good quant fund manager doesn’t look at alpha factors in isolation. By exploiting the way in which factors interact with each other, it’s possible to greatly improve success rates.
Consider the focus of our analysis so far: value factors. These factors interact with a host of other factors. For example, cheap stocks are more likely to outperform if:
Poor performance isn’t driving the attractive valuation (ie there’s an interaction between value and momentum factors)
The cross-sectional variation of analyst forecasts isn’t too wide (value and certainty).
There are no structural growth issues (value and growth).
Earnings and dividends aren’t being supported by unsustainable debt levels (value and debt)
We illustrate this point using the methodology we used previously whereby we adjust the manager’s level of skill for a long and short universe of stocks selected based on one year forward earnings yield in Japan. However, in this case we run an additional simulation whereby the stock selection methodology combines different alpha factors (Chart 3).
Chart 3: Value and Multi-Factor IRs Source: OQ
Both simulations will generate relatively low levels of idiosyncratic risk given the initial screening is based on a factor that is included as an independent variable in the regression analysis. Yet, the performance of the second simulation is clearly superior across all skill levels.
Ignoring the benefits of stock diversification
One thing I find interesting is the (increasingly) large number of managers who promote their high conviction investment approach. These managers are overrepresented at the top of league tables, as well as the tail end of the list. Hence, they tend to receive more than their fair share of publicity.
To provide a framework for evaluating this issue, we’ll use the Fundamental Law of Active Management: IR = IC * sqrt(Breadth).
Put simply, this formula states that risk adjusted performance (IR) is a function of the manager’s skill level (IC) and the number of independent bets the manager takes (Breadth).
High conviction managers with limited stock diversification require a high level of skill. It’s my view that high conviction managers tend to overstate their skill level given all the uncertainties associated with equity market investing. They may have generated great returns but as we discussed in a previous research insight, it’s very hard to distinguish between luck and skill for these managers. In some cases, it can require a two-decade track record to make this determination.
Conversely, a manager with a highly diversified portfolio like us doesn’t require as much “skill” to generate strong returns. In effect, so long as we get it more right than wrong, we can generate attractive returns.
To illustrate this point, we return to our methodology that involves an initial long and short screen based on one year forward earnings yield in Japan. In this case, however, we assume the same level of skill, but we adjust the number of positions in the portfolio.
Chart 4: IR Distribution (Concentrated vs Diversified) Source: OQ
Clearly having a highly diversified portfolio is an enormous benefit, yet it isn’t captured by idiosyncratic risk. Based on the methodology, all the simulations will have a relatively low level of idiosyncratic risk despite the extreme difference in the information ratios..
Overlooking the risks posed by high leverage
Our starting point for this analysis is that fund return volatility is required to generate decent headline performance. We’ve been able to generate an information ratio close to 2 using our discretionary multi-factor investment process at Macquarie Bank and more recently at our new firm. This is exceptionally strong. Yet if return volatility is low, say less than 5% pa, the fund return will be in single digits.
The challenge then is to generate decent return volatility in a way that minimizes drawdowns and the risk of any blowups. In broad terms, there are two approaches: take appropriate risk bets or employ high leverage. If the manager takes appropriate risk bets – or what we often refer to as informed risk bets - leverage can be relatively low. Conversely, if all risk exposures are heavily constrained, a high level of leverage is required. There’s no way around this; there’s no free lunch.
High leverage is not the answer. This should be obvious to anyone who remembers the LTCM debacle in the late 1990s, the Great Quant Unwind in August 2007, and more recently, the Archegos blowup in early 2021.
Yet, maximizing idiosyncratic risk (as a percentage of total risk) doesn’t capture the risks posed by leverage.
To illustrate this point, we’ll focus on LTCM. This was the granddaddy of leverage blowups given the US Federal Reserve was forced to intervene to prevent severe financial market contagion.
LTCM employed a relative value arbitrage strategy that endeavored to exploit small mispricing anomalies. The Nobel Prize-winning economist Myron Scholes described it as being akin to a giant vacuum cleaner “sucking up nickels from all over the world”.
The strategy had low exposure to traditional risk factors and no doubt looked attractive based on the level of idiosyncratic risk as a percentage of total risk. It’s the ultimate example of how a narrow focus on risk modelling while ignoring more important (and obvious) risks such as leverage is suboptimal.
Not analyzing country risk properly
We analyze country exposure based on “country of risk”. Although we believe this is a superior approach, risk models are based on “country of listing”.
“Country of risk” takes into account where a company sources its revenue, where management is located, and the reporting currency.
The distinction is most pertinent for China and Hong Kong. We have stocks in our China universe that are listed in China (‘A’ shares), Hong Kong (‘H’ shares), and the United States (Chinese ADRs). There are also several stocks listed in Hong Kong that have a “country of risk” located in Europe.
The relative pricing between ‘A’ and ‘H’ shares tends to mean revert and it’s a liquidity driven pricing anomaly which obviously we want to exploit. However, it greatly reduces our idiosyncratic risk as country of listing (used by Barra and Axioma) is only seeing a net long exposure to stocks listed in Hong Kong. It’s not offsetting this with our net short exposure to Chinese stocks listed elsewhere. Further, it’s not taking into account that some of our long positions in Hong Kong are actually European
plays.
We’re then left with a choice of doing something which makes sense in terms of generating alpha or something which makes sense in terms of maximizing our idiosyncratic risk for marketing purposes. For us this is an easy choice – we go with the former. We know that ultimately our success will be driven by the robustness of our investment process and our ability to generate alpha. This brings us to the next and final section of this research insight.
The final word … taking a smart and “differentiated” approach
As we don’t do anything purely for marketing reasons, we’re not held hostage by the level of idiosyncratic risk using a simplistic methodology. This means we can focus on the issues that really matter. Importantly, it also means we can take a differentiated approach.
In the investment world, it’s always good to be doing something different. The less money there is chasing the mispricing opportunities you’re seeking to exploit, the easier it is to generate alpha.
Regarding this issue, the backdrop is already favorable given there is less quant factor money in Asia than anytime over the last 15 years. Adding to the investment case, we believe a lot of the remaining quant factor money is being forced to dial down their exposure to important alpha sources for marketing reasons, making it easier for us to target uncrowded mispricing opportunities.
Our investment methodology and approach to risk management is based on many years of experience. In my case, I started doing Asian quant factor research in the mid-1990s. We’ve done the modelling, understand the pertinent issues, and have the requisite expertise.
Hopefully this is reflected in our research insights, including this one.For those who are interested, I also wrote a book on quant factor investing (please contact us for a copy).We’re passionate about our work and are very happy to share our views on a range of issues, not just the folly of focusing on idiosyncratic risk.
Comments