My last post on this generated some good questions. I’m going to answer them here, because this model deserves a better explanation. Before I start, I should say that in order to understand the model, you need to read the first two articles in the series, which are here:
If you are curious about the model, the information is there. It includes links to the main article at Economic Philosopher’s blog ( @jesselivermore on Twitter).
On to the questions:
Is this nominal or real return? Where can I find your original blog post explaining how you calculate future returns? Similar charts using Shiller PE, total market cap to gdp, q-ratio etc. all seem to imply much lower future returns.
This is a nominal return. In my opinion, returns and inflation should be forecast separately, because they have little to do with each other. Real interest rates have a large impact on equity prices, inflation has a small impact that varies by sector.
This model also forecasts returns for the next ten years. If I had it do forecasts over shorter horizons, the forecasts would be lower, and less precise. The lower precision comes from the greater ease of forecasting an average than a single year. It would be lower because the model has successively less power in forecasting each successive year — and that should make sense, as the further you get away from the current data, the less impact the data have. Once you get past year ten, other factors dominate that this model does not account for — factors reflecting the long-term productivity of capital.
I can’t fully explain why this model is giving higher return levels, but I can tell you how the models are different:
- This model focuses in investor behavior — how much are investors investing in stocks versus everything else. It doesn’t explicitly consider valuation.
- The Shiller PE isn’t a well-thought-out model for many reasons. 16 years ago I wrote an email to Ken Fisher where I listed a dozen flaws, some small and some large. That e-mail is lost, sadly. That said, let me be as fair as I can be — it attempts to compare the S&P 500 to trailing 10-year average earnings. SInce using a single year would be unsteady, the averaging is a way to compare a
outdatedsmoothed income statement figure to the value of the index. Think of it as price-to-smoothed-earnings.
- Market Cap to GDP does a sort of mismatch, and makes the assumption that public firms are representative of all firms. It also assumes that total payments to all factors are what matter for equities, rather than profits only. Think of it as a mismatched price-to-sales ratio.
- Q-ratio compares the market value of equities and debt to the book value of the same. The original idea was to compare to replacement value, but book value is what is available. The question is whether it would be cheaper to buy or build the corporations. If it is cheaper to build, stocks are overvalued. Vice-versa if they are cheaper to buy. The grand challenge here is that book value may not represent replacement cost, and increasingly so because intellectual capital is an increasing part of the value of firms, and that is mostly not on the balance sheet. Think of a glorified Economic Value to Book Capital ratio.
What are the return drivers for your model? Do you assume mean reversion in (a) multiples and (b) margins?
Again, this model does not explicitly consider valuations or profitability. It is based off of the subjective judgments of people allocating their portfolios to equities or anything else. Of course, when the underlying ratio is high, it implies that people are attributing high valuations to equities relative to other assets, and vice-versa. But the estimate is implicit.
So…I’m wondering what the difference is between your algorithm for future returns and John Hussman’s algorithm for future returns. For history, up to the 10 year ago point, the two graphs look quite similar. However, for recent years within the 10-year span, the diverge quite substantially in absolute terms (although the shape of the “curves” look quite similar). It appears that John’s algorithm takes into account the rise in the market during the 2005-2008 timeframe, and yours does not (as you stated, all else remaining the same, the higher the market is at any given point, the lower the expected future returns that can be for an economy). That results in shifting your expected future returns up by around 5% per year compared to his! That leads to remarkably different conclusions for the future.
Perhaps you have another blog post explaining your prediction algorithm that I have not seen. John has explained (and defended) his algorithm extensively. In absence of some explanation of the differences, I think that John’s is more credible at this point. See virtually any of his weekly posts for his chart, but the most recent should be at http://www.hussmanfunds.com/wmc/wmc161212e.png (DJM: the article in question is here.)
I’d love to meet and talk with John Hussman. I have met some members of his small staff, and he lives about six miles from my house. (PS — Even more, I would like to meet @jesselivermore). The Baltimore CFA Society asked him to come speak to us a number of times, but we have been turned down.
Now, I’m not fully cognizant of everything he has written on the topic, but the particular method he is using now was first published on 5/18/2015. There is an article critiquing aspects of Dr. Hussman’s methods from Economic Philosopher. You can read EP for yourself, but I gain one significant thing from reading this — this isn’t Hussman’s first model on the topic. This means the current model has benefit of hindsight bias as he acted to modify the model to correct inadequacies. We sometimes call it a specification search. Try out a number of models and adjust until you get one that fits well. This doesn’t mean his model is wrong, but that the odds of it forecasting well in the future are lower because each model adjustment effectively relies on less data as the model gets “tuned” to eliminate past inaccuracies. Dr. Hussman has good reasons to adjust his models, because they have generally been too bearish, at least recently.
I don’t have much problem with his underlying theory, which looks like a modified version of Price-to-sales. It should be more comparable to the market cap to GDP model.
This model, to the best of my knowledge, has not been tweaked. It is still running on its first pass through the data. As such, I would give it more credibility.
There is another reason I would give it more credibility. You don’t have the same sort of tomfoolery going on now as was present during the dot-com bubble. There are some speculative enterprises today, yes, but they don’t make up as much of the total market capitalization.
All that said, this model does not tell you that the market can’t fall in 2017. It certainly could. But what it does tell you versus valuations in 1999-2000 is that if we do get a bear market, it likely wouldn’t be as severe, and would likely come back faster. This is not unique to this model, though. This is true for all of the models mentioned in this article.
Stock returns are probabilistic and mean-reverting (in a healthy economy with no war on your home soil, etc.). The returns for any given year are difficult to predict, and not tightly related to valuation, but the returns over a long period of time are easier to predict, and are affected by valuation more strongly. Why? The correction has to happen sometime, and the most likely year is next year when valuations are high, but the probability of it happening in the 2017 are maybe 30-40%, not 80-100%.
If you’ve read me for a long time, you will know I almost always lean bearish. The objective is to become intelligent in the estimation of likely returns and odds. This model is just one of ones that I use, but I think it is the best one that I have. As such, if you look the model now, we should be Teddy Bears, not full-fledged Grizzlies.
That is my defense of the model for now. I am open to new data and interpretations, so once again feel free to leave comments.