“Hedge funds fail when rock stars are in charge”
Matthew Lyn, Bloomberg, 2010
This is the subject line for a Bloomberg news story I read in 2010. It resonated with me at the time, and it still rings true.
The news story referenced a research paper on how media coverage affects the performance of hedge funds. The research studied the performance of almost 1,000 hedge funds and correlated their performance against media coverage, measured by the number of mentions on Google News. The conclusion was that “hedge funds with media coverage underperformed no-coverage funds by 3.5 percent annually over 1999-2008”.
Four reasons were put forward for this finding:
Media coverage is backward looking.
High profile managers attract too much money leading to capacity problems.
Managers seeking publicity tend to believe their own hype and this can lead to hubris.
Some managers were more focused on publicity and raising money than managing money.
All these reasons make sense. I also find it strange that allocators believe that the person who is constantly marketing and drumming up publicity is also responsible for running the investment process. It’s just not possible. Managing a portfolio is a huge responsibility that requires dedication and commitment. It’s not possible to do this effectively while also being obsessed with self-promotion.
“If you torture the data long enough, it will confess to anything”
Ronald Coase, 1961
I’ve done a lot of backtesting over many years. It’s surprising easy to get good backtest results. For example, if you use different quant factors and weights for different stock groupings, you can get extremely strong returns. Adjust the factors and weights over time and the returns look even better. Often, it’s not necessary to torture the data, just strong coercion is required to yield the desired results.
Unfortunately, the results are only strong in-sample; the investment process is unlikely to generate strong out-of-sample (ie real life) returns.
I firmly believe that data mining (data fitting) is the biggest reason why many quant funds don’t live up to expectations. It’s tempting to generate the strongest possible backtest results and VAMI charts with the long and short portfolios moving in opposite directions look great in marketing presentations.
There are two key safeguards against data mining:
1. Out-of-sample testing.
2. Understanding the mispricing anomaly being targeted and why it exists.
Out-of-sample testing can be difficult when focussing on one market (given the need to time slice the dataset) but is relatively easy for us given our fund invests across Asia Pacific markets and we can run backtests across all key global equity markets.
The stock screens that comprise the systematic component of our stock selection process generate strong backtest results across markets. We run the same screens for each market and, except for financials (where different factors are required for some factor families), we run the same screens for each sector. We could get better backtest results if we tweaked the screens for different countries and sectors, but we do not believe these results will translate into superior real-life returns.
The second safeguard – understanding and justifying the anomaly – is potentially open to abuse. It’s possible to justify almost any backtest result ex-poste. Starting with the anomaly and then building the process with a strong focus on the mispricing opportunity being targeted is a good way to counter this. Focusing on behavioural biases which are based on human nature (and hence tend to be enduring and pervasive) is also important.
“High conviction, concentration, best ideas and assorted similar terms used with monotonous regularity in investment circles, profess certainty which doesn’t exist in the hope of conveying confidence to prospective clients.”
Martin Conlon, Schroders, 2023
This quote eloquently expresses my view on the breadth versus concentration argument. Many fund managers feel compelled to focus exclusively on their best ideas based on their perceived superior level of skill. Unfortunately, portfolio concentration makes it more difficult to determine the level of management skill and means (ironically) that the level of skill is less likely to matter.
This is a matter of simple maths.
To illustrate this concept, we use our favourite coin flipping analogy. Let’s assume you toss a coin with a 55% probability of landing on heads and the objective is to toss more heads than tails (in which case you win). This is akin to having a meaningful stock picking advantage over the market. If you only toss the coin 35 times, there is an 27.5% probability you will lose. Put differently, there is a material chance your “skill” isn’t reflected by the outcome.
In the real world, the outcome from stock picks isn’t binary and the marginal level of skill will likely decline when more stocks are added to the portfolio, but there is no doubt that concentrated portfolios make it more difficult to distinguish between luck and skill and make it easier for incompetent managers to obscure their shortcomings. For managers with highly concentrated portfolios, it can take over a decade of performance data to determine with a reasonable degree of statistical certainty the results are due to skill rather than luck.
If there is a significant style bias, it can take even longer.
There is also the issue of the number of stocks required to achieve the benefits of diversification. High conviction managers like to quote academic research studies on US equities which show that around 30 stocks are required.
As reported recently by Bloomberg, this finding has been debunked by Roni Israelov, a former principal at AQR and now chief investment officer for a financial services firm called NVDR. His research paper can be found here: https://ndvr.com/journal/how-many-stocks-should-you-own.
And the conclusion: you need at least 200 stocks to get the benefits of portfolio diversification.
“We were seeing things that were 25-standard deviation moves, several days in a row.”
David Viniar, CFO Goldman Sachs, 2007
This quote references the Great Quant Unwind in August 2007. Notwithstanding the fact a 25-sigma event is essentially impossible, it was without doubt a momentous event for quant factor managers, including me.
The Great Quant Unwind was a liquidity driven event. The initial catalyst was probably one large market neutral quant fund or trading book managed by a proprietary trading desk unwinding its positions. The exact details are unclear as no-one was able to pinpoint where the original trade unwinds came from.
This unwind created a cascade effect that ultimately spread more broadly to all quant factor funds and portfolios. The damage was done over 3 days between August 7 and August 9 – and it was severe. Quant factor funds suffered extreme losses over a few days that would normally only occur during a worst-case scenario over a few years.
Many quant fund managers either:
were forced to unwind positions due to margin calls (resulting from excessive leverage),
unwound positions as a result of systematic risk controls such as stop-loss triggers, or
overrode their investment process by reducing position sizes to mitigate losses and to allow them to re-assess the situation.
Fortunately, at the time we recognised the extreme price moves were due to liquidity distortions and, given our leverage was relatively low, we weren’t forced to unwind positions. Indeed, we traded opportunistically by running a short-term screen to exploit the price distortions and selectively added to long and short positions. Quant factor performance then rebounded on August 10 and drifted upwards over the remainder of the month. And best of all our fund registered a positive return during the most tumultuous month ever for quant funds.
“Nobody ever got sacked for buying IBM.”
Unattributed quote
Ok, this isn’t a finance quote but it’s pertinent, nonetheless. Risk aversion applies to all industries, including (and particularly) finance.
In finance, risk aversion manifests itself in both stock selection and manager selection.
In terms of stock selection, risk aversion is the reason why our turnaround screen exhibits strong predictive power. Investors are reluctant to embrace credible turnaround plays due to the reputational risks associated with buying price distressed stocks. A losing investment in Amazon will likely be viewed much less harshly than a losing investment in Alibaba. The thought of having to explain how you didn’t realise that Alibaba is at the mercy of the CCP’s regulatory whims is too much for many investors to bear.
In terms of manager selection, it explains why the biggest challenge for many hedge fund managers – except “rock stars” (refer to first quote) - is to onboard the first couple of investors. Returns are likely to be strongest when a fund is small and nimble, and will potentially enjoy the tailwind of additional fund inflows. In other words, it’s often a good to invest early, but the fear of making a losing investment in a small fund is too great.
Conversely, when a fund is large with numerous investors – and even better if it’s nearing capacity – investors are far more likely to show interest. If the investment goes badly, the reputational risks are much lower. And as the saying goes, misery loves company. In moments of despair, it’s nice to know that you’re not alone.
“Selling your winners and holding your losers is like cutting the flowers and watering the weeds.”
Peter Lynch, One Up on Wall St, 1989
This is a great quote on how a common behavioural bias can result in irrational decision making.
Related to this is the saying “it’s not a loss until you sell”. While technically true, waiting until
the share price exceeds the purchase price is not conducive to sound decision making.
It’s this type of bias which generates mispricing opportunities that can be exploited using a robust multi-factor quant investment process. The tendency of investors to sell winners and hold onto losers can be exploited using robust momentum factors that take into account the average entry price of investors.
“In the short run, the market is a voting machine but in the long run it is a weighing machine.”
Benjamin Graham and David Dodd, Security Analysis, 1934
This is an old quote but it’s still true and will always be true. Stock fundamentals drive long term performance while investor sentiment drives short term performance.
This is the reason why we include a broad range of quant factors in our stock selection process which have varying decay profiles. For example, we use value factors which exhibit relatively poor short-term predictive power but, given the predictive power dissipates slowly, are relatively effective at predicting longer term returns.
At the other end of the time decay spectrum, we also use short-term momentum and forecast revisions factors to exploit market and sentiment shifts. This enables us to generate alpha during different market regimes, including periods dominated by irrational exuberance (such as in the second half of 2020 and early 2021) when stock fundamentals and valuations are largely ignored.
“Picking up pennies in front of a steamroller.”
Quote attributed to various people, including Martin Wolfe and John Kay.
This quote typically refers to out-of-the-money option writing strategies.
I think it also applies to the use of extreme leverage for highly risk constrained investment processes. Extreme leverage combined with tight risk constraints works most of the time but is susceptible to market dislocations. Witness, for example, the demise of LTCM.
We also witnessed this phenomenon leading up to and during the Great Quant Unwind in August 2007. The quant funds used tight right constraints (which was good), but too managers were using the same alpha signals (which was bad) and hence it was necessary to use excessive leverage to generate decent returns (which was really bad).
The conditions precipitating the Great Quant Unwind are no longer present and hence it’s not a risk.
Potentially, the multi-manager firms which use extremely high leverage are susceptible to a severe market dislocation, but their diversified alpha sources and tight right constraints may protect them from an extreme drawdown. Time will tell.
Commenti