De Prado channels Marc

Looks like De Prado is saying what Marc has said all along: Three Quant Lessons from COVID-19

Three main points:

(I cannot count the number of times Marc mentioned interest rates—as does De Prado relating to #3 in De Prado’s talk)

  1. More nowcasting, less forecasting

“Traditionally, quant strategies have focused on forecasting prices, based on price time-series dynamics (e.g., stat arb), or based on cross-sectional data (e.g., factor investing).
The point in case is that days before the COVID-19 selloff started, there were plenty of warning signs that the virus was disrupting critical supply chains in China. This selloff may have been a Black Swan to market forecasters, but to market nowcasters, it was a predictable outcome. It is time for quants to pay less attention to crystal balls and add nowcasting to their arsenal.”

  1. Develop theories, not trading rules

"It is common for academics and practitioners to runs tens of thousands of backtests to identify a promising investment strategy.
The best performing backtest is then reported as if a single trial had taken place, and selected for publication or for launching a new fund. As a result of this selection bias, most published discoveries in finance are false, even if we cannot know exactly which. This fact easily explains why many funds have not performed as expected, including but not limited to the recent performance of many quant funds during the COVID-19 crisis."

3 Avoid all-weather strategies

“Academics and practitioners usually search for investment strategies that would have performed well across many different market regimes. That search implies the assumption that such strategies do exist.
But why would that be the case? Why would that source of alpha exist continuously, regardless of the underlying market and economic conditions? To quote Napoleon Bonaparte: “One must change one’s tactics every ten years if one wishes to maintain one’s superiority.” We all know what happened to him when he stopped following this advice.
Asset managers should focus their efforts on searching for investment strategies that perform optimally under well-defined market regimes.
“We hope that the quant community will adopt these three lessons. The COVID-19 crisis could help launch a new era of quantitative models that take advantage of more comprehensive datasets, are better aligned with the scientific method, and are more adaptive,” said the report.”

BTW, this is really James’ stuff (who used to post here and made this available to me by email).

Best,

Jim

Jim - This is a fantastic (and very brief) presentation that all quantitative investors should read. Thank you for posting it! It’s succinct, well-written, and makes some excellent points far better than I’ve heard elsewhere. It has convinced me that I have been wrong about a few things recently.

Yuval,

Thank you.

If anyone has been wrong it is me and as I imply in the “Post Topic” it is not just de Prado (but Marc also) that I have to thank for the learning experience.

Best,

Jim

Yes, this is a great paper. But the real lessons starts now: After reading it, APPLYING the principles laid out. That is the hard part.

Let’s start with this question. How do you characterize a “market regime”? Clearly, Lopez de Prado has interest rates as one essential component. What are others? Inflation would have to be one of them, I assume.

If we had, say, five parameters, we could look at various historical combinations to classify the past into different “market regimes” (and then see where we are today by comparison). We could then look at two “market regimes” that are separate in time but similar in characteristics and determine whether the same factors outperformed in both. If so, then we’d have a reason to allocate certain factors to certain “market regimes.”

But rather than starting from scratch, has anybody actually done this? Has Lopez de Prado, for instance?

Or is the best guess as to our current market regime simply the market regime that’s closest in proximity (e.g. the last three to ten years), and should we just stick with the factors that have worked during that period? That is my current approach, but I’d be glad to reconsider it.

I think Hedgeye has something like this. They have growth and inflation as “regimes” and depending if those 2 are either accelerating or diminishing, create “4 quadrants” or “regimes”. They claim they have backtested this and give recommendations on asset allocation accordingly, i.e. for each “quadrant” that they think the economy/stock market is in.
Has anyone tried their service?

There are some logical ‘categories’ for ‘regime switching.’

  1. Volatility (high volatility or low volatility regime). Can use arbitrary VIX cut offs for this, or look at VIX expansion or contraction. High vol environments favor stocks with larger market caps and greater financial strength generally (at least on the way down – which is a big part of returns) – lower vol or falling vol environment favor small growth stocks and momentum and mean reversion factors both (generally short-term mean reversion off stocks with medium to long-term positive momentum).

When volatility rises, bid-ask spreads break down, market makers vanish – price discovery is harder – and newer companies with less ‘earnings power’ or ‘dry powder’ and more need for growth get hurt.
2. Government interference in markets (have to manually code this or look at data proxies in some way) – high, medium or low. If high, look to sectors / companies favored or backtopped. Look for ‘historical factors’ to breakdown with more government involvement (i.e. short bans in China is past, or Ford / auto subsidies or FED price support). Generally ‘free markets’ stay free – unless major price dislocations – but some markets less free/transparent then others (i.e. China small caps much less free / transparent than SP500).
3. Market cycle. Growth phase / expansion phase (growth stocks beating general market returns) vs. contraction phases – looking at price series (are dominant ‘risky assets’ above SMA200 and/or are ‘risk free’ assets beating risky assets. Unemployment and GDP growth trends and CPI. If in a growth phase, may tilt towards momentum and higher beta stocks and stocks / sectors with growth stories – and lean less on valuations. Tech sector / Nasdaq outperformance and retail / consumer discretionary. If contraction phase will look at financial strength, stability and lower betas – and weaker (lower) momentum (more mean reversion) – and more ‘defensive sectors’ – healthcare, staples.
4. Panic vs. no panic markets (classifier) – When VIX spikes and all risky asset prices plunge – there are panic markets with severe disolations. What you do depends on capital deployed vs. earnings power and time commitment for investment till funds needed.
5. Inflation – high vs. low. If high, looking at companies / industries that can pass rising costs onto consumers and end users easily and/or with end goods / plants / capital equipment where revenues will rise faster than costs. Traditionally people look at miners, but can look at companies with very high pricing power and very short term contracts – with less price sensitivity (insurance, apts, maybe luxury goods, ‘sin’ industries, SaaS businesses in SMB space if their customers have cash still, commodity industries (if not in mining business).
6. Interest rates (low, medium or high). As interest rates fall, risky assets get (relatively) more attractive – more capital flows in. Dividend stocks from financially strong companies unlikely to cut those dividends also get more attractive (so long as the assets backing those payments aren’t linked to interest rates – so REITS portfolios may lose value in these environments, force div cuts).
7. Energy prices (price of oil in particular drives a lot of industries – airlines for example). So, when ‘oil shocks’ happen, tend to see airlines spike if they are just price drops, but fall if they may indicate geopolitical instability. But, this also feeds into 'volatility. May avoid energy sector or even certain energy utilities when energy prices fall.
8. Currency stability / geopolitical instability – but these feed into volatility. If a country was having currency issues vs. reserve country, may want to ‘avoid them’ – if you feel you can predict currency direction, should trade it directly.

Here’s a simple SP500 regime switch, monthly rebalance rank – using a few of these – built it after reading this question. It just have 5 factors total with 1 conditional node. So, not super optimized.


1., 3., and 4. are problematic. If you use volatility or growth as a measure of market regime then obviously those factors are going to work/not work depending on what regime you’re in. The only way we can use market regime as a guide to what factors we use in each regime is if we define the regime independent of stock factors. The other five measures, though, definitely hold some promise as the beginning of what could be an interesting experiment. The next step would be to break the last fifty years down into regimes according to these measures. And the step after that would be to see how value, growth, volatility, and momentum factors fared in each time period, and whether or not one could say with any consistency that such-and-such factors work under such-and-such conditions.

I just read an excellent and brief article that argues that the regime we’ve been in for the last five years is very far from over. See https://www.morningstar.com/articles/976372/such-a-narrow-stock-market

I have very little confidence that the industries and factors that have fallen into disfavor over the last decade will come flying back and predominate when this is all over. To me, the safest “regime” bet is the most recent one.

A perfect example of the need to read critically, now more than ever. That article was so grotesquely clueless, it cannot even be discussed. Forget his answers, he didn’t even understand the questions to ask. Be prepared for s lot more of this. When regimes changed, those who made careers under the old regime emotionally resist recognizing that things have changed and ramble on for years spouting babble that once made sense but has completely lost relevance and eventually, they wither into obscurity. Moving from one major regime to another is hard, and quants, the folks who most tightly bind themselves to the old regime, tend to struggle the most . . . and many big names today will gradually vanish from the scene.

Exhibit A. I rest my case.

Tomyani’s post is more forward looking. I might debate some conclusion here and there, but ultimately, he’s doing what needs to be done . . . Seeing what’s going on and thinking about how the core tenets of our political/economic/market sensibilities are likely to evolve.

We need to think big picture. . . .

“I have very little confidence that the industries and factors that have fallen into disfavor over the last decade will come flying back and predominate when this is all over.“

I wouldn’t even bother to agree or disagree with this. The question being addressed is so teeny tiny, nano and them some, it isn’t worth discussing. Opinions on what will work going forward have to be formed with ZERO consideration of anything empirical and COMPLETE focus on what’s going on in the world and what it may mean. Understand the world first. Then worry about factors.

Yuval,

That is probably the approach I will follow. Try to figure out what regime we have been in for the last 6 months.

Especially if I start to question the performance of my ports or ETFs.

The truth is I do not really have a prefect name for the regime we are in now although the Morningstar article you link gives me more insight. I am calling it the value and small-caps suck regime or just FUBAR for short.

I am not just trying to make a dumb joke, I do not have a name.

So mostly I need:

  1. An early warning that some thing has changed.

  2. Better skills at determining what regime we have switched to so I can know the smartest thing to do going forward.

How long to you think regimes might last? Do you think a goal of catching and adjusting a regime change 6 months in would be adequate/effective (generally)?

Edit So in practice I am doing what Marc recommends. I am looking forward thinking about what is happening now. But I might add that it is has been clear for more than 6 months that something has changed. I guess I am like Marc in that I want to add recent events (the virus) into my thinking. Recent evens are a bit unusual, however.

I guess I have to consider Marc’s points and de Prado channeling Marc’s points by looking at “nowcasting” closer. But that shouldn’t stop me from noticing when my ports have been giving me headaches and nausea over the last 6 months and thinking about that.

Best,

Jim

Marc,

“…Understand the world first. Then worry about factors.”

This statement is a bit too general to be of any practical value. What does it mean in concrete terms and what is following for devising a strategy? Can you elaborate?

Werner

(Try to wade through what may at first appear as philosophical BS. You’ll get your answer.)

Think of it as a two-phase process.

  1. I’ll steal the title of Yuval’s blog series (“How to be a Great Investor”) and boil the whole topic down to one sentence: Figure out what’s going on in the world and which companies are more likely to prosper from it. That’s it. All the great investors have done that. All those that haven’t have not done it at all or did it with lesser degrees of success. P123 can’t help yo with that. Neither can any other platform, quant article, etc. This depends 100% on your own human brain. Much of the academic financial literature reflects nothing more than a futile attempt to escape this core reality (and of course comply with university publish-or-perish mandates).

  2. This is where p123 and other platforms can step in. Once you figure out where the world is going and which companies are more likely to prosper, you need a way to give those companies a chance to identify themselves to you. (Notice I did not say give yourselves a chance to identify those companies. You can’t do that unless you study deeply on each company, learn the details of the industries, understand how management thinks, etc., etc., etc. You don’t have the time, and in the post Reg. FD era, if you managed to really do it, you’ll probably get indicted for insider trading. So don’t bother. You can’t identify them. They have to jump up and identify themselves to you. Your job is to put yourselves in a position to hear receive the information the companies are giving you; to hear what they are saying. Screening and ranking is th mechanism through which you can hear which ones are speaking to you,

Now, notice what I did not say.

I did not mention benchmarks, alpha, equity curves, drawdowns, or anything else like that. Stuff like that is for feedback, monitoring, etc. None of that sh** can help you invest better. And you need to interpret your results with extreme humility. If you haven’t figured out by now that past performance doesn’t indicate future outcomes, then there is no hope for you. No matter how many fancy graphs you construct, at the ned of the day, the market is in control, it does what it does and there isn’t a fu**ing thing you can do about it or to predict it, All you can do (1) is assess your own risk profile and tactically allocate between cash and risk assets accordingly, and (2) invest in companies that have told you they are well positioned to prosper from where the world is going and trust that over time, Mr. Market will reward their stocks, even if there are many periods, some severe, when Mr. Market is distracted from his main order of business.

And now, we’re coming down to factor, or in other words, or in other words, the language companies use to identify themselves to you through p123 as a language translator. For you and stock selection, it’s all about the factors I’ve discussed and the interactions between them in the virtual strategy design class, countless posts in the forums, and most recently, in the March 24 webinar and the recording I assume p123 is making available to you as well as the slide deck that accompanies it. No, t’s not about the same old factors. It’s about how you deicide to get them into your models. You get them into your models because financial theory shows you how and why those are the ones through reasonably priced which shares of companies prospering from whatever the current regime is push themselves into your result sets. So if you want details on factors, go to the March 24 webinar.

That webinar didn’t cover an important element; growth proxies. I don’t work for p123 any more so whether or not that next webinar takes place is up to p123. If p123 wants it, I’ll do it. If not, get it from the virtual strategy design course and I’ll my be addressing it on my personal blog (actiquant.com) and on the Chaikin blog (Chaikin PowerFeed - Chaikin PowerFeed); my blogging pace has been very very slow lately due to bandwidth constraints, but as some big Chaikin projects approach final implementation, I’ll be ramping up my writing. But my near-future blogging plans on Chaikin are likely to be heavily geared toward advisors and ETFs, so webinar or p123 written material is the best way to assure you get that content.

What’s really interesting here is that the factor strategy doesn’t need to change (side from whatever adjustments you may make to your risk threshold from time to time. Winning strategies don’t change from regime to regime. What changes are the kinds of companies that will appear to you. Another thing that changes is wether Mr. Market will pump those stocks in any particular period.

The hard part for you is to summon up the courage to sweep aside the bullsh** papers and articles you read, the p123 forums peer pressure to data-mine your way to brilliant equity curves while at the same time convincing everyone you aren’t really data mining, to withstand big drawdowns you’ll be able to predict a year or so after they end and you have time to refine your timing models to account for yet another new version of the past.

So, my suggestion to you is to listen to the Mar 24 webinar, read through the slide deck, and for more detail, go into the strategy design course. This is how you address task 2, hearing worthwhile companies that are trying to identify themselves to you.

Part 1, figuring out where the world is going, read and THINK about the news and what it MEANS. (You’ll need to do the latter on your own; don’t expect a reporter to have a clue and be able to explain it to you.) What you also want to do is reduce the role of the media middleman. Focus on company conference call. If you can get on open live streams great, do that. Otherwise, get the audio and/or written transcripts from company IR sites or seeking alpha. One call won’t do anything to you. Do lots and lots of them, for companies whose shares you own, for big high profile companies – you can never do too many.

Over time, you’ll develop a knack for seeing where the world is going long before any of it winds up being reflected in economic data and in the media. This is not a forever thing. It’s a training thing. Once you hone your sense of reading the world, you’ll be able to do more of it on your own, with help from some key paragraphs you’ll find in company earnings releases. And absolutely positively go to company IR sites and download company presentations and study with a critical, but not cynical eye. (Companies are sill allowed to portray themselves in a positive light, but they go to jail if they lie; train yourselves to glean relevant info without buying in hook line and sinker)

I’m happy to do webinars that focus on honing your practice in this sort of thing. Whether those happen will be up to p123.

Marc,
Wow, that is a hell of a detailed answer !!
Thank You.
Lot’s of pointers and things to think about (and things to do).

Werner

The webinar is here: https://www.portfolio123.com/app/webinar

And the slide deck, which is really nicely done, is here: Strategy Design Part 1.pptx - Google Drive (you can also access it from our “Library” page here https://help.portfolio123.com/hc/en-us/articles/360016297412-Library-and-Courses )

A little more data about this from Bloomberg: Quants Have a Fundamental Issue With Indiscretion

Made available to me by James

The more things change the more they stay the same? Is there any difference in how the market has fled into the arms of FAANG stocks from how it previously was drawn to the Nifty 50 during a late stage bull market?

[quote]
The more things change the more they stay the same? Is there any difference in how the market has fled into the arms of FAANG stocks from how it previously was drawn to the Nifty 50 during a late stage bull market?
[/quote]This! Don’t forget the dot com bubble, which was also a late stage bull market.

I think there is indeed. Facebook, Apple, Amazon, Google, and Netflix have all benefited to some degree from the stay-at-home orders. The large majority of other companies have not.

The Nifty 50 and the dot-com bubble were quite different. In neither case were there economic constraints that would clearly have hurt the large majority of non-Nifty 50 or non-dot-com stocks.

The FAANG stocks got a earnings boost from tax reform in 2018. In 2019 the tax reform was already priced in to the stocks, but they went up based on momentum. They are already overpriced and the fact that as a group they benefit more than most (in some cases temporarily), should not be a reason why their PEs should continue to skyrocket. However, that’s the nature of long bull markets. During the Great Depression, the largest stocks held up best, until the last year or so before the very bottom when their prices finally collapsed too.

As theory gives me headaches (I still wonder how I earned a PhD), I have put things in a very down-to-earth way.
I have looked at all my old screens and sims those:
1/ beating their benchmark since the model was designed in the US and in Canada (double out-of sample test).
2/ with a smaller drawdown than the benchmark in March.
3/ simple enough to be coded in a spreadsheet from data exported from other platforms
The idea of 1/ is to get a chance of eliminating curve fitting (but no guarantee)
The idea of 2/ is to find models with a low-vol profile when meeting fears of what the future market regime could be (even if we don’t know what it will be). Low-vol might not be the most powerful anomaly, but it is likely the most stable through history.
The idea of 3/ is being independent of the platform. I love P123, I trust the team and hope it will last forever, but I won’t bet my retirement’s money on that.
I have a few candidates (no stellar returns expected), but I need to have a closer look and make more tests with the new Factset data.