If you are reading this, you are hopefully well aware that here at IvyVest, we believe in taking a systematic approach to investing. We like to adjust our allocations based on the cold, hard, numbers, rather than by relying on “gut feel” or intuition. So we tend not to react to the day to day “news” of the market as much as many others. We believe that the financial news is “noise” more often than it is “signal.”
However, being systematic about investing does also should not and does not mean locking ourselves away in a news-proofed chamber and letting black-box algorithms do all of the heavy lifting. In fact, we think that a purely statistical approach to investing can be as equally dangerous as a purely “gut-feel” approach.
A big problem with taking a purely statistical approach to investing is that you can become highly susceptible to what is called “over-fitting.” This is a fancy way of saying that it is easy to come up with a way to explain what has happened in the past, but much harder to predict what is going to happen in the future.
The danger of over-fitting is especially pronounced when we have a large amount of factors that we could potentially use to explain things relative to the amount of historical data that we actually have on those things. For instance, with the financial markets, we have maybe 80 years of solid historical data, at best. This is actually less than it seems relative to the number of things that could have happened in that time. It would be nice to know, for example, what would have happened to asset prices if Hitler had won World War II, or if the Russians hadn't turned around the boats during the Cuban Missile Crisis. But we never will.
Meanwhile, we have an almost unlimited litany of potential “indicators” that we can use to explain what has happened in that past. Something is bound to fit, even if it is only by chance alone.
One way to avoid the danger of over-fitting is to only use factors in the model that actually make economic sense. For instance, it could be that there has, in the past, been a seemingly statistically significant correlation between the occurrences of sunspots, and stock market returns. But it doesn’t make any sense at all that sunspots should have anything to do with stock market returns. It is a lot more likely that this is just a random correlation. So we don’t include any data on sunspots in our model. This is obviously an extreme example, but the point is we can guard against over-fitting by only using factors that we think should work based on an informed view of how the markets operate.
A more pernicious form of over-fitting can occur if the future is a totally different environment than the past period that we built our model on. This can be an extreme danger for those that try to generalize lessons from a very narrow sliver of history. For instance, if you had been living in 1980 and had informed your investing approach entirely based on what happened in the 1970s, you would have likely built a system that would have probably performed very well in a time of rising inflation (such as the 1970s). Unfortunately, over the next thirty years, the prevailing trend was of falling inflation, and your model likely would have under-performed. Of course, many people who invest intuitively based on their own life experiences are committing this exact error -- since those life experiences accumulated during a relatively short time period that may be unrepresentative of the future.
We partly guard against this danger by a) Using as long a “training” data set as we can (we have data on some markets and factors back into the 1800s) and b) explicitly considering the economic environment in our model by splitting assets depending on how they perform in rising or falling inflation and on how they perform in rising or falling economic growth. We make sure to diversify across all groups.
Even so, at the end of the day, designing a systematic investing framework is as much of an art as a science. It ultimately relies at least somewhat on judgement on the part of the model creator as what to what factors will and will not be important in explaining asset returns in the future (which is why we take pains to explain our philosophy).
A critically important step in the process is to periodically “stress test” that judgement by imagining all the ways that it could go wrong in the future. We should constantly be asking -- how might the world be changing today such that what we have built based on the past may no longer work in the future?
This entire stress-test process would be too wonky and long to go into here in its entirety (and I also don’t want to give all of our secrets away), but I thought it could be useful to open up the hood a little by looking at one particular aspect of the model: how we decide how richly priced US stocks are. This is a key factor that helps determine how aggressive or defensive the final portfolio allocation is.
The valuation of the US stock market is also a particularly relevant and timely thing to be looking at, because some commentators have been arguing that we are entering another "bubble." Others contend that we are in a secular bull market that still has room to run. As always, we would rather let the data be the judge.
As we have written before, a useful starting point to looking at stock market valuations is to compare the overall level of an index (such as the S&P 500) to the total earnings that all companies in that index generated in the previous year (weighted by their overall market capitalization). This is known the historical price-to-earnings ratio (P/E). The good thing about this measure is that we have relatively good data on it that goes back for some time, so we can get a good idea of where the market is at today relative to its history. Here is what the P/E of the S&P 500 looks like over time:
On this measure, the market looks like it is only now slightly moving ahead of its post WWII average (represented by the dotted red line). Things by no mean seem out of hand yet, and the "bubble" calls look very premature.
However, this is not the actual measure that we use in our model, nor is it the one that the “bubble” crowd is citing. What we (and many of the bears) look at instead is a measure that was studied by Noble Laureate Robert Shiller, a professor at Yale. Shiller measures the valuation of a market by dividing price by an inflation-adjusted average of earnings over the past ten years. His point is that earnings can be volatile from year to year due to economic recessions and booms, but that by averaging over a sufficiently long period of history we can get a smoothed series that better predicts how companies are likely to perform in the future.
Here’s how Shiller's cyclically-adjusted PE measure looks today, relative to the same post-1950 period:
We are still far below where the market went in the 90s tech bubble, but are starting to creep into "overvalued" territory.
What explains why the stock market looks reasonably valued on forward-looking price-to-earnings basis, but expensive on more normalized measures? Here’s the explanation, in a single chart:
This chart shows the profit margins of US companies over time (expressed as a percentage of the overall economy, using data from the Federal Reserve). The key takeaway is that companies are more profitable as a whole today than they have ever been in the past.
This matters because using a cyclically-adjusted price-to earnings ratio is implicitly assuming that profit margins revert to their average level over the last ten years. As you can see from the above chart, this has generally been a good assumption to make over the past few decades. Empirically, profit margins have mean-reverted whenever they approached their current level. Logically, this reversion to the mean makes sense given what we know about economics. High profit margins should incentivize firms to grow and hire more workers, which then drives wage rates up (and profit margins down). And vice versa.
So where could we be wrong by making this assumption? It takes a little bit of thought, but it turns out that asking why profit margins have increased is equivalent to asking why wages have stagnated. This is because profits are corporate revenues that flow to shareholders (owners of capital), while wages are corporate revenues that flow to workers.
This is a question that economists have been very interested in. Popular explanations fall into two categories: 1) Outsourcing and 2) Technology.
Of the two, outsourcing has gotten the most political play, probably because it’s always a good political idea to have someone to blame for taking “our” (speaking from a US perspective) jobs. In particular, the entry of some 600 million Chinese laborers into the world economy may have been a one-time “supply shock” that held wage rates down and allowed companies that already had the know-how, distribution relationships, and supply chains intact to make windfall profits by substituting cheaper workers in Asia for expensive ones in the US and Europe.
However, as this new source of labor becomes assimilated into the market, Chinese workers will grow wealthier and start to demand consumer goods in addition to producing them. Both production and wages will eventually increase, and the old relationship may start to reassert itself. And indeed, there are signs that this process is already occurring. Wages have been increasing in China, and the pool of “surplus labor” seems to be drying up, at least somewhat.
It is the second explanation - technology - that could in theory produce a more significant and lasting change in the relative returns of capital (what we call profits) and labor (what we call wages). Recently, several prominent economists have started to argue that we are entering what they have called a “second machine age” -- a time when machines start to replace not just manual labor, but when improved machine-learning algorithms can also displace traditionally “white collar” analytical tasks. If what “capital” (machines, factories, and intellectual capital like patents that are controlled by companies) is capable of doing is rising relative to what “labor” is capable of doing, then it makes sense that the returns to capital would rise relative to the returns to labor. And, more importantly for our purposes, this shift could be at least semi-permanent. Put differently, if we live in a world where robots are doing most tasks, then it makes sense that a greater share of the pie will go to the owners of those robots (provided that the robots haven't discovered how to unionize yet, at least).
Of course it is also true that many of these profits may accrue to start-up companies that have yet to go public (or, in many cases, yet to be even be created), so it is not totally clear how much the companies in the current stock market may benefit. Nonetheless, this theory is an interesting one to consider.
In conclusion, a thorough testing of our approach to valuing US stocks a key potential limitation that we take seriously. If profit margins remain elevated due to a step-change in the technology vs. labor tradeoff due to “the second machine age”, then our long-term measure of smoothed profits may be too conservative. If this were the case, our suggested allocation to US stocks might be lower than it optimally should be. On the other hand, if corporate profits were to revert to their historical average, then stocks could quickly start to look more expensive than any of us currently imagine (since our measure is based only on the last ten years). Under this scenario, stocks could go from a 20 P/E to a 40 P/E without any change in price.
While we think it is important to acknowledge these uncertainties, we are also relatively comfortable with them. We would rather error on the side of history repeating itself, rather than making the bet that this time will be different. Most of the time, things that have mean-reverted many times in the past continue to mean-revert. Only occasionally, they do not.
So while we are not making any changes to the model as a result of this “stress test”, we are keeping a close eye on what is going on in the economy. Given a change in the facts, we will happily change our opinions, and change our model. Anything less would be insanity.
We've written in the past about the problems with relying on past return measures when evaluating new investment opportunities. This month, we tackle a bit of a more straightforward subject: how to calculate those past return numbers. As with most things in investing, it's not quite as straightforward of a matter as it first looks.
Three Kinds of Returns
In fact, reading the financial news on any given day, you might encounter three different kinds of returns (and they won't always be labelled to let you know which one is which).
1) Price Return. "Price" returns are calculated from market prices, and therefore ignore dividends. Why does it make sense to ignore dividends? Well it doesn't, really, but if you are a reporter running late for an assignment, it's often a lot easier to do so, because you can just take quote data directly from the stock exchanges. Price returns can also nefariously show up in some places that you might not expect them. For example, the S&P 500 Index that is commonly reported on in the news, is a pure price index -- so calculating returns based on it will totally leave out dividends.
2) Total Return with Dividends. The second way to treat dividends is to add them into the return figure at the time that they occur. This is the same as assuming that dividends flow into a money market found or bank account (or are spent) at the time that they occur.
3) Total Returns with Dividends Reinvested. The final way to treat dividends is to assume that you use them to repurchase additional shares of the investment. For an investment that goes up a lot over time, this can generate significant additional returns, since it enables this money to "compound" and grow exponentially.
Annualizing Returns: Simple and Geometric
Making matters more confusing, the returns that you see reported for periods of longer than a year will frequently be "annualized." This is because it is more convenient to think in terms of how much you will make per year. It's a common unit of account.
Because returns compound exponentially, you cannot just take the total return over a period and divide by the number of years in the period to get the average annual return. This would be very wrong. For instance, a 50% return over 5 years is not a 10% annual return, it's a about an 8.4% annual return (1.5 raised to the 1/5th power - 1).
This compounded annual return is also known as a "geometric average" or a CAGR (compounded annual growth rate).
The other way that annual returns are often reported is a simple or arithmetic average annual return. This is calculated by taking a straight average of all of the annual return figures for an investment. If you had a data table of the calendar-year annual returns for a stock going back to 1960 and you took an average of every figure in the table, this is the number that you would get.
It's important to recognize that the simple way of calculating average returns will yield a different figure than will calculating a geometric average return. Geometric average return is generally the figure that you should be interested in. And of course, both of these numbers can also vary depending on how dividends are treated.
At IvyVest, we almost always will use compounded annual returns with dividends reinvested. We think that this is the most realistic figure for an investor contemplating actually holding a strategy.
The moral of the story? The return figure isn't quite as simple as it seems. Know what you are looking at before you jump to any conclusions too soon.
|Period||Model Return||Benchmark Return|
All models have been updated with the latest market data and can now be viewed on the portfolio page or from the cloning tool. Since the market has not moved that much in the last month, we are not recommending that you need to rebalance this month. A "rebalance" alert will be triggered whenever the model changes by more than a threshold that we have set.
Get our next article delivered to your inbox.
Sign up below and be the first to know about our freshest data-driven thinking on the markets, and investing. We will send you no more than one email a week. This is free.
Ready to start putting this into action?
Take a free two-week trial to IvyVest premium -- our premium subscription service. You'll get access to our rules-based dynamic asset allocation model, tools that will show you exactly what you need to buy in your own discount brokerage account (and when to re-balance) to implement it for yourself, and an insightful monthly newsletter that will keep you on abreast of the most important things going on in the markets. There is no credit card required. Get Started Now!