Back to Community
2019 equity quant themes underperformed - why?

Anybody have a root cause analysis of Fawce's observation:

Through 2019, in the face of challenging market conditions, many equity quant themes underperformed as valuations continued to rise.

I'm not looking for speculations about Quantopian specifically, but rather what about the equity market would have led to "equity quant themes" underperforming (presumably, he's talking about pure alpha strategies, versus pure beta or so-called "smart beta").

The implication is that something unanticipated happened at a macro level, but what? Fed actions? Tax cuts? Politics? Economic outlook? Etc.

36 responses

That's a good question, but I’m not sure anyone has a good answer. Here’s Cliff Asness recent perspective on it (or related subject anyway - the Value factor getting cheaper and the spread wider compared to historical).

My personal (and far from unique) take on it is that one reason quant factors do well in the long run is that at certain (shorter to medium terms) they (unpredictably) perform poorly and appear to have ‘stopped working.’ If something worked well all the time (e.g. whatever the secretive RenTech Medallion fund does), everyone would start doing that, and it would stop working. Therefore (in my view anyway), quant factors under performing unpredictably from time to time is actually a good thing, and a key reason why they continue to outperform in the long run.

Following Joakim's observation, those principles of occasional underperformance in the shorter run also hold true in traditional value investing and were discussed at length by Ben Graham. If a certain method always worked, everyone would do it.

I don't have the figures on hand, but it looks like Q has performed unremarkably since inception based on the total payouts to authors. I wonder if this is caused by some overlooked factor or overemphasis on a high Sharpe without truly supporting high frequency trading. With enough entries, you're bound to get lucky for a period of time (in the contest) but eventually that winning streak runs out if the principles of the algo aren't valid.

Does anyone have insight into what this means for Q users, beyond fewer algorithm constraints? I would certainly welcome a good quant environment for foreign (S. Korea, Japan, Taiwan) equities or bonds.

With enough entries, you're bound to get lucky for a period of time (in the contest) but eventually that winning streak runs out if the principles of the algo aren't valid.

100%. Law of large numbers!

The strongest gains in all regions were found among long-short equity funds, which returned 8.64%.

Q had been running a long-short equity fund, and 8.64% return is pretty decent, but I guess if SPY returns a lot more, you have investors chasing beta returns...but then isn't that the point of a hedge be a hedge against beta?

One hypothesis is that share buybacks caused a disconnect between models and the 2019 reality. If you have an arbitrage scenario where companies can either borrow money against their good credit and/or use windfall tax breaks to reward their shareholders, then what's the model for pure alpha? In hindsight, one would need a company-by-company buyback detector (I'd assume that the tax breaks more-or-less lift all companies uniformly).

Share buybacks, FEDs hidden QE4, historically low interest rates and late stage of bull market cycle when euphoria kicks in is the cause of Q fund underperformance in my opinion. But then again, maybe our strategies arent good enough yet.

@ Vedran -

Yeah, that's my hypothesis. How does one make a forecast when the market is driven by some combination of government intervention and irrational exuberance? Maybe Q should host a challenge to write algos that would perform well when the party is over?


How does one make a forecast when the market is driven by some combination of government intervention and irrational exuberance?

I have no idea. I know that I wouldn`t have confidence in that kind of model that easily.

Maybe Q should host a challenge to write algos that would perform well when the party is over?

I agree with you. Sad thing is that "when the party is over" algos struggle in backtest (mildly said) before the party is over.

I've seen algos that trade on Fed announcements and sentiment. There was a forum post here a while back that just inversed StockTwits and generated some pretty solid alpha. Sadly, that edge has since decayed meaningfully.

One could also factor a fundamental, qualitative thesis (GSIs will be bailed out in the future, and therefore have a built-in put option) into an algorithm (don't short large-cap American banks, regardless of the signal). I do a similar thing in a non-Q algorithm I run, albeit not with that example.

If your algorithm generates real alpha and doesn't rely on providing liquidity to markets, it ought to perform in all environments. If it fails to (on average, of course) in either today's market fueled by financial cocaine or during a global crisis, your alpha isn't actually alpha. You just failed to account for certain risks.

The obvious example here is LTCM. Funds like it appear to be generating real alpha in the intermediate term, but are actually just implicitly short volatility/tail risk, whether through the actual methodology or though leverage.

LTCM just confirmed that even Nobel prize winners are greedy.
As far as "real alpha" is concerned I think we have quite good factors here just waiting to be used.
I have tortured sentiment data many times and unfortunately ended up with low alpha and huge portfolio turnover as result.
If someone created promising scalable algorithm which uses Fed announcements and sentiment please share the results.

I very much agree that we have a lot of good factors that do generate "real alpha." It's just difficult to differentiate that from a purely quantitative standpoint until SHTF.

It wouldn't work terribly well on Quantopian, but I imagine one could develop any number of strats in forex, short-term bonds, interest rate swaps, or the like based on breaking news or underlying economic data, let alone doing some sentiment mining of Fed minutes. We just don't have strong enough infrastructure (or market data, for that matter) for the more exotic financial instruments often involved.

Also, see here:

Great discussion.

I have a number of long-short hedged strategies that have performed fantastically during 2019/2020, so I think it's a cop out to blame the difficulties on long-short or a rising market.

Of course data-mined balance sheet correlations can always simply turn out to be spurious -- this must be a part of the problem. Typically backtests will be inflated (curve fit) at least to some degree.

I also do agree that Trump bullying the Fed, PPT, low rates, "not"-QE, corporate tax break, etc. change the market calculus. Many of these are examples of one-offs. How do you account for uncharted territory in an algorithm? There's no way to make evidence-based decisions for market regimes that have never occurred before. An expert human money manager with a deep understanding of financial markets could deduce and make informed projections, but an algorithm cannot -- it wouldn't be evidence-based. I'm assuming all these algo strategies were trading as if history simply repeats instead of factoring in this uncharted macro-economic territory.

While interest rates, etc. explain why money has been pumped into the market.. a rising sea lifts all boats, which hedging should neutralize. However, this has not been the case in 2019. So something else is also at play...

Fawce's post makes it sound like the fund was way too overweight variations on "value," which expects undervalued stocks and overvalued stocks to mean revert in terms of valuation. During 2019 fundamentals did not improve, so all of the market's gains were due to "multiple expansion." In other words, instead of valuations reverting, they became more extreme. As interest rates are so low, I think that's the proper market behavior, to invest in the Amazon model of reinvestment and delayed gratification. You can ultimately create more shareholder value that way, so long as the business is a killer.

The tail risk that occurred was multiple expansion. I'd argue that this wasn't even a wildly unlikely scenario. Had the Q Fund had more style diversification, perhaps it would have fared better.

Of course, there's also always alpha decay as the market becomes more efficient. In recent years we've witnessed the "value" space more and more filled with so called "value traps." I'm talking about Sears, GE, etc. Algorithms that bet on their recovery were just ignorant, focusing on balance sheet numbers instead of the actual business. In other words, just like growth stocks deserve high multiples (because of their discounted future cash flows), many cheap stocks deserve to be cheap.

@ Viridian Hawk

I agree with pretty much everything you said. Still, for some reason, people always think that this time is different and disregard fundamentals and when SHTF Warren Buffet somehow appears with his 160 billion $ in cash.

Sub-par Programs Tend to Underperform

If a trading strategy does not perform as expected, the reason is not necessarily that the nature of the market changed or whatever anomalies that existed were arbitraged out. No single individual or group could arbitrage the market. It is simply too big. If, and it is a big if, if they could succeed as a group to successfully arbitrage the market, then stock prices would barely move. And it is not what you are seeing almost every day of the week.

I would venture that the main reason for failing to outperform would be programming for something that might not have been there in the first place or was simply coincidental over some past data. In other words, making a program and expecting that the future will be like the past. You can give it any moniker you want. It has been known since the late '60s that the majority of money managers underperform the markets (see, for instance, Jensen's 1969 paper). And this, even when the majority has the needed brainpower and access to all the latest tools (hardware and software).

Making assumptions on passed and delayed market-related data which might have little relevance to what will be coming is also another way to underperform. In a sense, saying that whatever predictability was assumed in that trading strategy was not that predictive after all. What is coming has never been seen. It is all new territory. There has never been 9 billion people with all their needs on this planet, and that is where we are going over the next 30 years.

A win rate of 51% is not distinguishable from strict randomness. A trading strategy not outperforming a benchmark is sub-par indeed. Stock prices tend to almost follow a martingale, and this implies that the expected price change will be zero. This should be viewed as no expected price change, no expected profit.

Therefore, the first thing to do is demonstrate that you are not dealing with a pure martingale, that your trading strategy has an effective long-term edge. Otherwise... you have nothing and playing on market noise. Nonetheless, you can play market noise and still win.

Developers tend to not take responsibility when designing sub-par programs and present over-fitted backtests which at first glance might show some promise based on the short-term trading methods used, but still have a poor strategy architecture. As if trying to escape the harsh reality of CAGR degradation over time either by ignoring it or not even considering compensating for it.

They treat trading as an investment vehicle when a lot of it might just be biased gambling based on some criteria which might not even have been proven to prevail over time, especially going forward. Typically, they will take credit for a backtest showing above-average returns, but will not take responsibility for other people blowing up their account using their program.

The concept of underperforming, especially using an automated solution, is deeply anchored in the stock trading strategy itself, and there, you are the designer and it is your program that is running your show. So guess what, maybe the program is sub-par, its premises and assumptions, factors, clusters, sectors, or what have you might not really behave as you see them. But this can happen anytime you digitalize a seemingly analog input having a high degree of randomness or uncertainty.

Those designing outstanding programs (outperforming market averages) will not make them public. I have not seen any hedge fund giving away their best programs. And I do not think I will see any of them doing so going forward. However, sub-par programs abound.

Quantopian announced that they have had over 12 million backtests performed over the years and some 6 million trading strategies including variations. Out of it, only a handful have been considered worthy of an allocation. So, how many sub-par programs are we talking about?

A win rate of 51% is not distinguishable from strict randomness.

Really? Wouldn’t that depend on the sample size? RenTech’s Medallion fund appears to have done quite well on ‘just’ a 50.75% success rate. Most casinos do as well.

A trading strategy not outperforming a benchmark is sub-par indeed.

Which benchmark are you referring to? SPY? If so, comparing market neutral strategies to a long only benchmark is not an ‘apples to apples’ comparison unless you adjust for risk. So yeah, on a risk adjusted basis I might agree. (Maybe)

Just to throw in another monkey wrench (I’m in a destructive mood today), if a strategy consistently performs ‘sub par’ there’s value in that too. Just long the benchmark and short the ‘sub par strategy’ and you’re done.

No single individual or group could arbitrage the market. It is simply too big.

The federal government is big enough. The U.S. stock market total capitalization is $34 trillion, and the federal government has a debt of $23 trillion. The idea that the market is "efficient" is false when one includes the influence of the federal government effectively creating arbitrage, and funneling money to shareholders. Since the financial crisis, there's a case that the federal government has created a market-wide arbitrage, and continues to do so (or the debt-to-GDP ratio would be going down instead of continuing to rise).

Is the SHTF moment now? Should we start filming Big short 2? Are value algos finally back?
Time will tell but damn this is going down fast and will need many rescue tweets.

Fear of the Corona virus impacting manufacturing in China? Or is the market feeling the Bern?

I think that both fundamentals and technicals aligned today.

Fear of the Corona virus impacting manufacturing in China? Or is the market feeling the Bern?

I think today is about Covid-19 hitting Italy, which means containment measures aren't working.

Over the past 5 trading days, SPY and GLD moved in more-or-less opposite directions (chart). The former is down 4.3% and the latter is up by the same amount. Kinda interesting. FUD and anti-FUD.

Gold = safe haven

Brief article on supply chain impact of coronavirus:

I suspect that companies don't even know what to say yet regarding the potential impact (at least quantitatively), as things unfold.

The idea of the Fed coming to the rescue is interesting. Ray Dalio has been warning for a long time that next down turn will be harder for the Fed to manage. Could get interesting.

Manufacturing companies which have the best relations with suppliers will get ahead :)

@ Vedran - I suspect it is more complicated than that. It only takes one item missing from inventory to shut down production. Building a car is a lot different from building a hammer, for example.

True. It is much more complicated than that but I am looking forward to see how can FED`s monetary tricks stop this systematic biological and now world real economy issues. They are not caused by monetary policy so I think that drastic FED decisions could even propel further downturn despite their best intent.

drastic FED decisions could even propel further downturn despite their best intent

Perhaps. It could be interpreted as an indication that more is needed than just to "Keep Calm and Carry On".

I came across some data on the Secured Overnight Financing Rate (SOFR):

Note the huge spike in mid-September 2019. Does anyone know why it happened? And did the Fed just head off another such spike this past week?

Some discussion:

FED injected over 50 billion dollars of liquidity. I would imagine a similar chart for the recent headline news too.

FED Meeting Minutes 9/17/2019

@ Daniel -

I'm not sure there was a spike in the SOFR this past week, but something weird happened:

"These changes are being made to address highly unusual disruptions in
Treasury financing markets associated with the coronavirus outbreak,"
the bank said.

Something was going on well before the coronavirus, if they had to intervene with $50 billion last September. Fishy.

I hope your signals were corona-resistant.
If the data/signal is good enough it will withstand most market conditions.

Loading notebook preview...

I doubt the 2019 Golden State Raptors, looked at Celtic's games from the 70's and 80's to gauge the current competition and devise a game plan going into the 2019 championships. While still significant, historic factors are antiseptic and lack a competitive ethos (my basis is purely intuition). Do quants embrace war? because the market is a battlefield. Sure Stanley McChrystal probably read Lao Tzu but I'm sure the 4 star JSOC General assigned more weight intercepting/interpreting enemy transmissions and building liaisons with insurgent warlords when laying out tactics than subscribing to an ancient Chinese Strategic philosophy (for instance). There are market players out there trying to juke your algo's, knowing your algo is running on stale notions.