Kub's Den

Correlation does not equal causation; we all know that. But as grain market participants prepare for another monthly round of information and projections from USDA's World Agricultural Supply and Demand Estimates (WASDE), it may be helpful to get out of the habit of expecting linear relationships of any kind between prices and those projected numbers.

For instance, it would be misleading to assume that a 10-million-bushel drop in livestock feed usage will "cause" a 10-cent drop in the price of corn ... and not the other way around. Furthermore, whatever the causative relationship between these two variables -- maybe lower feed usage causes prices to drop or maybe higher prices cause feed usage to drop -- there may not be any linear relationship between the two variables whatsoever. With no linear relationship, there is no way we could expect a meaningful correlation between the two sets of numbers.

Let's ignore the thousands of other variables that are all changing at the same time and confounding the data. Let's say we could really simplify the world and somehow assume from one day to the next the value of the U.S. dollar won't change, the price of oil won't change, the price of fertilizer won't change, the expected April snowfall across the Corn Belt won't change and the hedge funds won't start suddenly selling or buying grain futures for some unrelated portfolio rebalancing reason, or anything like that. Let's assume the only thing that will change between Feb. 8 and Feb. 9 will be one number -- the feed-and-residual-use projection on the 2023/24 Corn Supply and Use table from USDA. Let's further assume we have 30 years' worth of monthly data and each time, these were the only two variables that changed -- feed usage and price. Even with that large sample size and perfect control of all the other variables, we STILL might not find a linear relationship between the two variables, to say nothing of proving which one causes the other.

A linear relationship could be something like this: If feed usage is 30% of the total supply, then the justified price of corn will be $5. If feed usage is 32%, then the price of corn should be $5.10. If feed usage is 34%, then the price of corn should be $5.20. And so on -- 10 cents higher for each extra two percentage points, or 10 cents lower for every two points diminished. Obviously, the world doesn't work this way because there are so many other simultaneous factors affecting the price, but even if we wanted to describe the effects of just this one variable, the relationship could take some other shape than a straight line. As of the January WASDE report, corn's 2023/24 feed and residual use number was projected at 5.675 billion bushels, or 34%, of the 16.727 billion bushel total supply. This was a lower proportion than the previous marketing year when 5.486 billion bushels going to feed represented 36% of the total supply.

The relationship between these two variables from one year to the next -- or from one month to the next or from one moment to the next as traders anticipate changes in the industry and incorporate fresh information into their trades -- might not be a straight line. It could be a parabola, or "U"-shape, created by a quadratic function showing that each time prices drop by y cents, usage increases by x2 + x million bushels. Or vice versa. Or whatever. It could be an exponential relationship, with the rate of change itself growing larger and larger as the independent variable grows larger. It could be an up-and-down, up-and-down sine wave. We don't know. But in any case, it's not helpful to have any expectation that the price of corn will change in some predictable, easy-to-understand, linear way just because we see the number on a table of economic projections change once a month.

In fact, an entirely different kind of reasoning may provide a better explanation for some grain market movements. Looking for statistical correlations of linear relationships is an example of causal reasoning, the process of identifying the relationship between two changing variables when one causes the other to occur. Causal reasoning allows us to diagnose what has happened in the past and make predictions for what will happen in the future. Our brains love to find (or invent) causal, linear correlation relationships that make the world seem more predictable and less scary.

What if the grain markets don't behave like that, but instead behave according to an entirely different style of reasoning -- for instance, like a complex series of if-then computer programming statements? A different type of reasoning, conditional reasoning, describes how conclusions can be drawn from a conditional proposition, like "If it rains today, that guy's hair will be wet," or "If the feed and residual number drops by more than 10 million bushels, the price of corn will drop by at least 5 cents." That conditional statement, especially if taken alongside a complex web of similar conditional statements describing every other factor that may be working on the market at the same time, allows for considerably more nuance than any blunt expectation of a linear relationship. Perhaps the feed and residual number will only drop by 5 million bushels.

From the conditional statement, we couldn't infer any conclusion about what might happen to the price of corn. Maybe nothing will happen. Or perhaps the price of corn will indeed drop more than 5 cents after the new supply and demand numbers are released. Observing that the "then" part of the "if-then" statement is true doesn't actually allow us to conclude anything about the "if" part -- in this case, the feed number. That's like an observation that the guy's hair is indeed wet -- maybe it was raining today, or maybe he just took a shower.

A third type of reasoning that could describe market price movements would be comparative reasoning. This happens a lot in the grain markets, whether we realize it or not. Think of when an old timer at the coffee shop says, "This winter feels just like the winter of 1988, and we all know how bad that turned out." Or an analyst says, "When we see China start buying soybeans from Brazil in these quantities, we can expect a weakening basis like we saw the past two springs."

In both cases, the speaker's brain is making a comparative connection between two or more scenarios and, upon finding certain factors to seem similar in the scenarios, is predicting what will happen next. In the vast computers running machine-learning algorithms (artificial intelligence), this style of comparison would be done by clustering algorithms. The "intelligent" machine would partition all the things it can observe about various scenarios, calculate which scenarios seem to resemble each other most closely (usually by lots of simultaneous linear regression and correlation calculations, but not necessarily) and then make predictions for how the new scenario at hand will behave in matching ways. Again, it can be a more nuanced way to predict the blunt assumption that one variable causes another variable to change in a linear way.

All of this is just a caution to anyone interested in the grain markets -- don't get too confident about the way our human brain likes to simplify the world into easy-to-understand linear relationships. Even changes to a very helpful statistic like the stocks-to-use ratio, which gives one nice overarching look at how leftover inventory (ending stocks, at 2.162 billion bushels of 2023/24 corn in the January WASDE report) compares with overall usage (14.565 billion bushels, or therefore a 14.8% stocks-to-use ratio), can never be used to make confident, linear predictions about what correlated prices will do from one month to the next, let alone from one year to the next.

**

Comments above are for educational purposes only and are not meant as specific trade recommendations. The buying and selling of grain or grain futures or options involve substantial risk and are not suitable for everyone.

Elaine Kub, CFA is the author of "Mastering the Grain Markets: How Profits Are Really Made" and can be reached at analysis@elainekub.com.

(c) Copyright 2024 DTN, LLC. All rights reserved.