Our new Constitution is now established, and has an appearance that promises permanency; but in this world nothing can be said to be certain, except death and taxes.
— Benjamin Franklin, in a letter to Jean-Baptiste Leroy, 1789
Franklin, if alive in today’s political environment, would feel vindicated. Alongside death and taxes, he could have listed one more invariant feature of our world: Year-end prophesy. This refers to, of course, our ritual forecasting of the Dow Jones Industrial Average performance.
At EidoSearch we dutifully participate in this ritual and once again put our stake(s) in the ground. With no further ado: One year from now, on Dec. 31, 2018, the price of the Dow has a 70% chance of falling within the range of 21,561 (down 11.6%) and 30,920 (up 26.8%) with an expected value of 26,284 (up 7.8%). That’s a mouthful, but any less info would be grossly inadequate—untestable prophesy—compared to the task of proper forecasting.
As Franklin so aptly observed: “…in this world nothing can be said to be certain.” We live by that motto. Lest we forget, the Dow could easily fall outside the bounds even of this wide 21,561 to 30,920 range. To be explicit, there’s an equal chance (15% on both sides) that the Dow sinks below or rockets above these boundaries. When we forecast, our stakes in the ground form a distribution, marking off all possible prices and their respective probabilities.
What’s the source of these probabilities? In one word: conditions. There are times in the past that barely resemble conditions today – throw those out. Global Financial Data, a premium vendor of market history, provided monthly prices in the Dow going back 130 years. That gives us a set of roughly 1,500 (12 x 130) returns to consider as we look ahead. Not all returns are created equal. We keep the times that are most similar to create a subset of possible outcomes in 2018. That’s the essence of our methodology.
Specifically, we found 232 periods from history that matched the Dow’s monthly movements over the last 12 months in 2017. None of these periods is a perfect match, but they all exceed a similarity threshold of 85%, and we further filtered by the magnitude of increase. We pulled out three matches that illustrate that range of potential outcomes.
The “Most Positive Match,” (right) with a correlation of .88 occurred from February 1927 through January 1928. The following 12-month period produced a rally of 59.9% in the Dow. The “Most Negative Match,” (below) occurred in a similar period, from October 1928 through September 1929. We all know what happened next. The Dow dropped more than 20% in the following October and 40.3% over the next 12 months.
We included a “Sideways Match” (below) that occurred from June 1970 through May 1971. That period had a correlation to the most recent 12-month period of .95 and the following 12-month period produced a return of 5.8%.
While it is clear that similar performance periods to that of the last 12 months have produced disparate future performances, you may take it as a cautionary note that in all three examples, a major market crash occurred within 20 months of that period. The Great Depression hit in October 1929 (20 months after the most positive match and immediately following the negative match) and the Dow made a major top in January 1973 (18 months after the sideways match) before dropping nearly 50% over the next two years.
How did our 2017 forecast pan out?
At the end of 2016, when the Dow stood at 19,763, after what felt like a death-defying multi-year march upward, there was a lot of rumbling among our brethren of the market losing steam. We predicted it would more likely gain steam. We forecasted a 70% probability that the Dow would land within the upside-skewed range of 24,549 and 18,650. As we put the final touches on this article (Dec. 12) the Dow stood within this range at 24,386, and more than 20% higher than where it left off in 2016.
So we were right? Well, it doesn’t quite work that way. When you make a single distribution forecast, in a sense, you can’t be shown to be wrong or right. Whatever the outcome – up, down, sideways – it fits somewhere in your probability distribution.
However, at EidoSearch we make thousands of systematic forecasts each day, and we keep track. It’s like counting pennies in jars. Listen for a "ding" each time you get a realized outcome in the Dow (or whatever you’re forecasting), and over time as your jars fill up, you can give yourself an accuracy score.
Say we make a sequence of 100 forecasts in the Dow – one each week for two years. We expect these conditional forecasts to contain information which, at its best, would mean this: 70 of the realized outcomes in the Dow should fall in our middle range (the big jar), 15 should fall above and 15 below (the two little jars). We could have divided the boundaries differently, but let’s save that topic for another time.
The world is uncertain. Forecasting is hard. Holding yourself accountable is even harder. But if you resolve to do it right, you will likely improve…and you will certainly pay taxes.