Since these tools aren’t widely used by laypeople yet, there isn’t a lot of public data on how well these models perform for commodity crops. In the post, The Chances of Sub-$3 Corn (And How to Reduce Them), we described how these models work in theory; in this post we’ll examine how well they can actually predict the future.
To help you understand how all this works, let’s start with a poker analogy. If someone told you the chance of drawing a three-of-a-kind was about 2%, how would you verify that? You could play thousands of hands with your friends, counting the times a three-of-a-kind was drawn, and you’d get a number close to 2%. Sounds easy! It turns out we can calculate the odds of $4 corn in much the same way: for each day of market data, we make predictions on the final price of corn and compare that to the actual final price.
Unfortunately, doing this isn’t as easy as playing a night of poker. With poker, the deck never changes. There’s one fundamental difference with crop prices: since the price of corn changes every day, the odds of $4 corn change with it. To deal with this, we’ll change our focus from a specific price to a specific risk level: in other words, we’ll answer the question, “when our model says the chances of something is 10%, how often does that actually happen?”. The chart below can help you visualize how these predictions look over time.
The blue line – the “Predicted 10% Risk” – shows one of our price predictions over time. If the final price is above that range close to 10% of the time, we’re getting the outcome we predicted. It’s like we’re drawing as many three-of-a-kinds as we expected we’d get.
But this chart shows us only one year of predictions. It’s like trying to guess our odds from a few hands of poker – we might not draw a single three-of-a-kind and think our odds are 0%. We need to validate this over many years. Over the long term, if the final price is above the 10% risk line close to 10% of the time, this suggests the model is pretty accurate.
We ran the same analysis for every corn future from 2001 to 2014 – a full 14 years of trading data. Just like we did for the 10% risk level, we assessed our accuracy for other risk levels, too. How often were prices above our predicted 5% risk level? 6.3% of the time. What about our predicted 20% risk level? Prices ended up above it 20.2% of the time. How well did the model fare for corn across all years of data? Surprisingly well. We summarize the accuracy of our model in Figure 2.
|Predicted Risk Level||Measured Outcome|
The folks over at Farmdoc have turned this model into a tool that answers our original question: what are the odds you’ll have a profitable season? Enter a price and you’ll see the chances that corn will end below it. If you can assess your chances of a profit, you can make more informed decisions on crop insurance, hedging, and other mechanisms to limit risk. Learn the 3 Practical Farm Risk Management Strategies from our previous blog.
Use these tools to make more informed risk management decisions – plan your farm like a trader would.