2018 – The Last Dance

Introduction

To the readers of this blog, I should perhaps introduce myself before I begin my first post. I am a casual retail investor, who happens to have a background in statistics. My interest in investing started a few years back from a simple question of “how should I invest my earnings?” This was especially pertinent then, because I was finally no longer making meager graduate student pay. Since then I’ve been spending an ample amount of ‘outside-of-work’ time learning about investing basics, and a nagging desire has been gradually building to share some of my educational adventures. The hope would be to provide a useful tidbit of information to the readers, and if not, then at least bring some clarity to my own thoughts through the writing process. This blog may later expand to include other topics, not directly investing related, as I happen to stumble upon something that I find impactful or interesting. Lastly, for those who have been patiently waiting for me to take this initial plunge into writing (or for those whose ears hurt from hearing me talk about this desire, but not act), now I can finally say, “I’ve begun!”

2018 Market Prediction

I felt a great first topic would be to share my prediction of how the investing world will turn out in 2018. To be specific, my primary focus will be on the outlook for equity markets and in particular, the S&P500 index (an index tracking the ‘top’ 500 large- and mid-cap U.S. stocks), as this index is most relevant to casual U.S. retail investors like myself. As a forewarning, none of the discussion below should be construed as investing advice; to be realistic, out of all the predictions from great investors and analysts out there, my prediction – one from a self-proclaimed newbie – is more like a decaying straw of hay in a large haystack (rather than the prophetic needle we all would like to find).

There are multiple ways to approach the markets, and I find for myself that a quantitative approach often appeals most to my senses; numbers often feel more concrete for me as I can play with them to reach a better understanding. I’ll also try to always remember to ask, “why am I wrong?”, because only by searching for what we are missing, can we start to be prepared for when our predictions do miss.

The heart of my prediction stems from a great post in a fantastic blog, titled ‘The Single Greatest Predictor of Future Stock Market Returns‘. In the post, the author describes how to construct a predictor of future 10-year returns of the S&P500 index (SPX), that beats most of the other commonly used measures, and has a outstanding correlation of 0.913! The essence of the predictor is to capture the degree of allocation to equities by investors, where the higher the allocation the lower the future 10-year returns, and the lower the allocation the higher the future 10-year returns (key chart is shown below; x-axis is time in years, left y-axis is average investor equity allocation, and right x-axis is future SPX 10-year total returns annualized).

avginv

The author does a great job breaking down why this predictor makes sense. Intuitively, the idea is if most investors have already piled in to the market, then it would be much harder to find additional investors to keep pushing-up the market. The above chart illustrates that this phenomenon does actually occur. If you follow the red-line, which represents the investor equity allocation, you’ll find that when the red line is low historically, subsequent SPX annualized 10-year total returns, as indicated by the blue line, tend to be elevated. For example, in 1991, investor allocation was ~25%, and the 10-year returns from 1991 onward through to 2001 (which would be found in a chart of the actual SPX prices) increased at an annualized rate of ~17%. In contrast, in 2000, investor allocation was at it’s peak around 52%, and the resulting annualized total return from 2000-2010 was only around -1%. Since the red-line aligns quite closely with the blue-line throughout, this means that investor equity allocation was highly correlated with observed subsequent 10-year SPX total returns (at least since 1952), and may potentially serve as a good predictor of future equity returns.

I would venture a guess that few investors invest with a 10-year horizon, and even fewer would be able to buy-and-hold SPX through that entire period. Thus, what might be  more helpful is if we were able to transform the above chart into a plot of observed and expected prices of SPX by year; visually this would more appealing as well. To do so, we would simply need to use the predicted total annualized 10-year return to calculate the expected SPX price 10-years in the future, based on the current SPX price for a given year of interest: (current price)*(predicted annualized return^10). And voila! you would get the chart below, where the black-line is the predicted price and the red-price is the observed price (historical data from here).

priceplot

We can see that, holistically, the SPX price version (shown here) of the SPX 10-year annualized total return chart (shown previously), does match reasonably well to observed SPX prices. In particular, I highlighted in blue boxes, a few periods where the predicted price did not change: (1) 1971-mid 1977, (2) 1999 – 2009, and (3) mid-2017 to 2024. Looking at the historical price change of SPX during those first two periods, they both contained major market crashes – the bear market of 1973-1974 where the Dow Jones Industrial Average lost over 45% of its value, the popping of the dotcom bubble in 2000-2002, and the most recent financial crisis of 2007-2008 – netting investors possibly huge losses in both periods. Extending this to the third period that has yet to transpire, this plot would seem to suggest that a similarly poor return is forthcoming, with a market crash (or crashes) some time during mid-2017 to 2026.

One important point is that from year-to-year, the observed SPX price may actually deviate quite substantially from the predicted SPX price. To better visualize this, we can plot the %difference over time:

perdiff

 

The mean (median) %difference ends up being close to zero: 1.9% (-0.01%), suggesting the overall prediction on average performs well, but for a given year the prediction could underestimate by as much as ~80% and overestimate by as much as ~-30%. This may seem like a much larger range than expected, but this is largely due to the scale of the original rate prediction compared to this price prediction; differences between the predicted and observed annualized 10-year rates are magnified due to the ^10 scaling to get to a price prediction.

Does this mean our SPX price prediction chart would not be useful? Au contraire! Aside from providing a general sense of when equities will tend to do well vs poorly in the future (ie, predict bull vs bear markets to incorporate into a more general investing framework), we can still use conservative adjustments to mark important price points to watch for. For example, if we take the ‘high’ point price predicted over the start of the third period in 2017-2018, this would correspond to SPX around 2800 at the end of 2018. Then we may want to make a -20% underestimate adjustment (based on the recent % differences suggesting consistent underestimation in recent years) to arrive at a peak price of 3360. An even more conservative (or liberal, depending on your point of view) estimate would be use -30% (largest underestimate historically observed) to arrive at a peak price of 3640.

Lastly, it is always important to ask what assumptions have I made that might cause me to be wrong? The most obvious risk would be that this relationship will no longer be true, i.e. historical predictive value may not translate into future predictive value, for a variety of reasons, like the environment may change in the future (as Irving Fisher had famously proclaimed, “Stock prices have reached what looks like a permanently high plateau” right before the 1929 market crash). Another concern may be whether the predictive relationship found was due to over-fitting to the data (i.e., torturing the data until it tells you what you want to hear). To address that concern, typically one might break the historical data into test and training periods, and/or support any observations with a logical mental model or explanation. Another idea may be to combine this model with other quantitative models for shorter time frames, or assessments of the macro ‘picture’ (won’t discuss here, but o boy! was it elucidating to learn about the killer D’s of debt, deflation, and demographics, excessive volatility selling, never-ending passive ETF inflows, etc.) to develop several supporting lines of evidence. One simple practical plan would be to simply watch for a change in the price trend to confirm any predictions.

Summary 

My prediction for the S&P500 index, and the U.S. equity market more generally, is the end of the current bull-market near the beginning of 2019 with a peak price level of ~3360 (with an estimated ‘max melt-up’ to ~3600 and an estimated ‘early stoppage’ at ~2800). An accompanying market crash is likely, with no end in sight yet to this next bear market (currently, at least through to 2027). This means that you’ll likely still have one last chance to dance your heart out before the party ends, but be careful of staying too long, as you might get stampeded by the crowd when the music stops, the lights turn on, and everyone rushes to go home. And you’ll finally get to clearly see who you’ve been dancing with.