It’s not just VALUE that does not work. Factors are just a gimmick dreamed up by the fund providers to sell their ETFs to the unsuspecting public. Check the Vanguard factor ETFs for proof of poor performance.
The most idiotic offerings are the multi-factor funds - why not just buy a fund tracking one of the major indexes if one wants more factors to invest in.
I can assure everyone that quantitative analysis is alive and well! You simply have to look at Inspector Sector’s Cloud Computing to come to that understanding. You have to run with “what works”. I have been criticized for making this statement before but it is now clear. What is working is NOT traditional value metrics, but other factors including revenue growth, smoothness of revenue growth, sales surprises, etc.
My guess is that if you want to understand what is going on then you would have to go back a hundred years to the last industrial revolution (assembly line) and see how growth stocks performed in that environment. That is where we are at now with digital transformation.
There are a number of reasons why traditional value metrics may not be effective. The first being that earnings is a comfort metric that investors had keyed in on. The truth is that Amazon has demonstrated that earnings are essentially a meaningless construct. Companies that declare earnings pay taxes, so why do that? Why not reinvest back into the company instead? Earnings may impress investors but what do investors really get out of it (earnings)?
One thing that I recently discovered is that Stock-Based Compensation increases the common equity which (I believe) results in higher book value. Another thing that I recently discovered is that deferred revenue for X-as-a-Service companies is often considered either current or long-term liability. It tends to show up as debt on the balance sheet. So these issues probably render the P/B ratio somewhat meaningless, at least for s/w companies.
P/S ratio is also not meaningful, especially when dealing with very high-growth stocks that are expanding into greenfields. You can throw P/S out the window as well.
In any case, the “state of the art” valuation is discounted cash flow valuation which superficially “works” but unfortunately every analyst on the planet misuses it. FOr example, what is the terminal growth of Apple? This company has been around since the late '70s and if one applied the standard analyst formula of estimated GDP growth, Apple would be growing at 2% per year, which clearly isn’t the case.
This brings me back to quantitative analysis. Correctly implemented, you extract the factors that are working in the present markets and run with them. How you determine what is working is the black art that you have to master.
Steve
Should note also that Cliff Asness says AQR’s proprietary value based metrics haven’t been performing well for 2 years. And even Renaissance Technology’s Institutional Alpha Fund (note, not the Medallion Fund) is down -21% for the year. Just glancing at their F13 they’re holding a lot of big pharmaceuticals like Bristol Myers Squiq and Novo-Nordisk which are obvious value tilts. So the struggles of value are not just confined to the obvious price based metrics that any Joe Schmoe can pull up in their Yahoo Screener.
These are good insights, as usual, thanks Steve.
A question though… if we are in the middle of a new industrial revolution, wouldn’t this show up in all global markets? Instead the growth seems to be mostly rewarded in the very developed world, and the US in particular. Seems to be a mixture of techology and loose monetary policy. It will be interesting to see if emerging markets experience a wake up when they start developing monetary policies that resemble developed markets (or if US growth declines once it no longer has that relative edge over the rest of the world).
Or this could trigger actual inflation, which would buoy commodity-rich Emerging market value and impair developed market growth
Interesting times
I’m not sure I understand the question. The beneficiaries are s/w companies, regardless of where they operate. Consider MELI and BABA, certainly not in the USA. The pandemic has dramatically accelerated the shift to the cloud. It isn’t loose monetary policy although that doesn’t hurt.
Inflation will not be a factor so long as baby boomers are retiring and interest rates remain low. i.e. people won’t spend if they are not earning substantial interest on their savings.
Steve
Actually, Steve nailed it. Don’t go ga-ga over quotes from gurus (something I see often in the forums, over passages and articles that range from mediocre to absurd to obsolete). In substance Steve’s points are far superior to those of the guru quoted at the start of this thread. If you’re not looking at and understanding what is and isn’t working . . . you’re DOA no matter how beautiful a set of backtested equity curves you can produce.
Around the time I ceased to be a p123 insider, I recall having posted and discussed at a webinar the areas that were working and anyone who took those remarks seriously and factored them into their modeling should have been doing very well. (I haven’t looked at the numbers for Steve’s Cloud Computing DM but just knowing he was hitting that theme is all I need to tell me it has to be looking sweet.)
P123 has Business taxonomy commands. Don’t be afraid to use them. You can’t do all math all the time. And interestingly, from what I can see, the FactSet taxonomy in areas that are working, and likely to continue working for a while, are quite strong . . . easily strong enough to blow away the impact of whatever other factor availabilities or PIT concerns some might have.
As to value, the factor sometimes works and sometimes doesn’t. But value “investing” always works. Big difference. A first-grader in the top 98% of his class can tell which P/E. P/S. etc. ratios are lower than others. That HAS TO tell you there’s more to value investing than just knowing how to count. (The quant gurus y’all love to quote never did get past a first-grader level of dealing with value, which is why they’re fu**ing up now and scrambling to try to understand why. For more, see Value “Investing” Always Works Even When The Value “Factor” Falters – Acti-quant which, actually, is an expansion of material I already handed you on a silver platter in the On-Line Strategy Course. I had a mini-Twitter debate with Cliff Asness on this but he politely – which is way out of character for him – ran off when it became apparent that we were, as he put it, speaking different languages.)
Steve,
Isn’t the success of your Cloud Computing model to a significant extent a sector bet that has worked? I’m not sure what quant-related conclusions I can draw from it.
Roger
To determine what is working now, is it possible to determine factors with the strongest momentum in the previous 6 or 12 months and then use the strongest 2 or 3 factors for a simulation?
Certainly, the model benefits significantly from the sector. But the point is that the sector has characteristics that are independent of other sectors such as energy, materials, or healthcare for example. You wouldn’t apply general-purpose value metrics to cloud computing, would you?
There are many different approaches to determine “what is working”. I choose factors that have performed best over the last 5 years and combine into an optimal RS. I use the RS for the next year. Rinse and repeat.
Personally, I wouldn’t use less than 12 months history, but in any case I suggest a 4:1 or 5:1 ratio, [In-sample]:[Out-of-sample]. So if you are using 12 months optimization backseat, then don’t run your model for more than 3 months before you re-optimize.
Steve
Further to this discussion. The latest RS for Inspector Sector’s Cloud Computing was developed back in November using the previous data vendor. So it has more than 6 months OOS under its belt.
I have just run the optimizer, swapping in different RS’s with a 6-month backtest. Obviously this testing is done with FactSet instead of S&P. I have also turned off preliminary data. There will be differences between what the Design Model actually achieved and what the backtest achieves due to the differences in data vendor and data fixes.
With this in mind, the original RS developed in November with different data vendors still comes in 2nd for %Return over the last 6 months. out of 15 RS’s. See attached.
The point is that the optimization process did a pretty good job. While it would have been nice if it had come out the best in the latest 6-month period, I don’t consider that to be realistic. There will always be some degradation over time with changing markets. The fact that the RS remains near the top speaks for the process used.
Will I change to “Core: Sentiment”??? I’m not sure yet. I don’t have a good feel for how reliable the backtest results will be, especially with sentiment data. Any thoughts would be appreciated.
Steve
jsk,
I am not sure I have a good answer to your question. But you may find this article interesting: Factor Momentum and the Momentum Factor
Best,
Jim
This is why we need a proper walk forward optimizer, imo.
I kind of simplified things by declaring victory based on OOS %return. In practice, I optimize based on the monochronic increase in RS buckets. I’m not sure that this would be automatable in real life.
Steve,
Your way is a fine way to do it. Indeed, I just funded a system and I optimized it in a way that cannot be called significantly different from what you describe. Not that my system is any good but I agree with the method.
But walk-forward, the way Korr 123 describes it, really belongs in a different category than my P123 model and perhaps yours too.
It is a way to at least try to adapt to a changing market regime. It can adjust to changes in the market. It is not optimized for the full 20 years at P123. That is not to say that I think it always works. Indeed, I would not argue if someone said it is difficult to make it work.
He means to use it in a dynamic way and changing way. Like looking back 3 months, plug that data into a regression, neural net or whatever and predict the next week’s returns.
Next week look back 3 months again (advanced 1 week from the last time you did it) and predict the following week.
In other words walk it forward each time.
Actually, the majority of my money uses such a system (from another site).
That does not stop me from using a system optimized in a similar way to what you describe here at P123 (I do).
But I do not think the two can be compared. Not apples to oranges but rather monkeys to clouds. Not one better than the other, necessarily. Just different tools for different situations.
And Korr is right that it can work, imo.
Here is a link to a McFarber paper as an example of a simple walk-forward model: Relative Strength Strategies for Investing
Many have read this paper and can form a quick judgement (or may already have an opinion from this paper) of walk-forward methods on their own.
Best,
Jim
Jim - I have dabbled in mechanical system development for a good 25 years. What you are describing is one very specific “walk-forward optimization” and I would be very careful regarding the results. Meb Faber is probably a bright guy but “A Quantitative Approach to Tactical Asset Allocation” was touted as a very strong approach to asset allocation and was tested in many different markets. However, it really didn’t work well out-of-sample.
Automated walk-forward optimization doesn’t refer specifically to relative strength but can be any set of factors. And I believe that it is important to establish the ground rules for selecting the best factors and their combination. It is in no way apples-to-oranges. If you don’t establish some common-sense ground rules upfront then you just end up with another hack at extreme optimization. It becomes very convenient to hit a button and watch the computer chug away printing out spectacular results that are simply wishful thinking. If you don’t get the results that you want then make an adjustment and repeat…
Steve
Steve,
I think you are right about that. In fact the possibilities are literally infinite.
Indeed, I think jsk was referring to a possibility outside of relative strength. Something I think is worth investigating.
I think Korr123 probably was not referring to relative strengh either but I am not sure.
Having only tried a few walk-forward optimizations myself I am hesitant to disagree with Korr123. Maybe he has experience with something that I don’t. Almost certainly he does.
There are enough papers for people to form their own opinions on relative strength (if they have not already). I was just illustrating that people have already (probably) seen an example of a walk-forward method.
Best,
Jim
Everyone here on the site optimizes. There is not a single person on this site who has not “thrown out” a backtest. How you optimize is based on technology (or lack thereof) and the quality of your research. Every successful quant fund optimizes. Citadel optimizes, RenTec optimizes, etc, etc.
Walk forward just optimizes based on the most information available at that time. I don’t really see how or why this is controversial. I hope that explains my remark.
P123 already tried and discarded AI because it “didn’t work”. I’m simply trying to avoid that situation again by suggesting that the criteria for selecting an optimized system may more complex than going with the best %return. At least we should have the option for making it more complicated than that if walk-forward optimization is ever implemented by P123.
Steve
NEW QUESTION: OPTION BASED. So I very frequently buy very long naked calls. Sometimes the cost of the option is so high I will spread by shorting an higher strike call against my long at the money call to lower the total risk exposure (net premium) but of course if the underlying stocks is a long term home run the higher strike call accretes value very fast.
Here is the question; in the above situation do you have a strategy as to when, if ever, to cover the higher strike long call?
Mike
