Cocoa Edition: Just how good is our platform at predicting prices?

Welcome to blog #2 in one of my favorite series, “Just how good is our platform at predicting prices??”  Following up on our awesome win as this year's Agri Business Review Commodity Company of the Year, we wanted to continue to demonstrate to our readers and customers why we’re the best in the world at what we do.  

As a reminder, the question we’re looking to answer is: If you had been using our signals over the past 10 years, how accurate and profitable would they have been?  In this blog post, we tested the predictiveness of Helios’ climate risk signals against Cocoa prices, to identify the optimal long only trading strategy.  There’s been a ton of money made this past year in the cocoa trade, but we wanted to make sure our signals were accurate AND profitable over the long term - which is why we’re showing you the data for the past ten years.  

Key points for the backtesting:

• Helios risk signals are leading indicators for price - the higher the risk, the more likely it is that prices will rise (i.e., a higher climate risk signals a higher likelihood of crop disruptions)

• Every trade was for a single contract - we did not increase or decrease the size of our bets based on the strength of the risk signal

• We initially looked at a “long only” strategy, where we would enter a trade after the weighted risk signal surpassed a certain level (y-axis, or rows) and then exited when it passed back below another level (x-axis, or columns)

As you can see in Figure 1, our risk signal is consistently predictive at all risk thresholds, and becomes even more accurate the higher it goes.  This means that for the vast majority of all trades undertaken, the risk signal is a leading indicator of price increases and a trader would make money.  We would note that the weighted risk for Cocoa has only gone above 45 six times over the past ten years, which makes those trades more rare.  

Figure 1: Accuracy of trading signal (i.e., how often did this trade make money?)

Figure 2 quantifies the absolute dollar return of each of these trading strategies in a ten year period, calculated as the total dollar return of buying and selling a single contract.  For example, every time the weighted risk went over 45, a single contract was bought, and then exited when the signal dropped below 10.  

The sum of all of these trades over a ten year period was $227,300, which was calculated by adding the profit/loss of each individual trade over this period.  For example, if the weighted risk signal went over 45 we would have purchased a contract at its full price.  Let’s assume it was $7,980.  When the risk signal dropped below 10, we would have exited this position.  Let’s assume the price then was $7,990.  The “profit” from this trade would have been our exit price ($7,990) minus our entry price ($7,980), so $10.  We then added all of the profit/loss outcomes for this particular trade strategy over the last ten years, to come up with the $227,300 figure.  Note this is absent any implied leverage (i.e., we assume in this strategy that we paid the full amount for a single contract).  As expected, the absolute returns are markedly higher given the dramatic increase in cocoa prices over the past twelve months. 

Figure 2: How much money would have been made in a ten year period? 

As we’ve said before, our platform should predict the price of soft commodities.  We have built the most comprehensive system of climate risk signals ever (14M+ locations; individual ML models for each commodity), which are often the primary driver of commodity prices.  The importance, and strength, of climate risk signals will only continue to increase as the impacts of climate change worsen.  Already we are seeing greater price, and climate risk, volatility than ever before.  

So where do we go from here?  Similar to the rise of quantitative trading algorithms, climate risk and artificial intelligence are becoming table stakes for institutional commodity traders.  So if you’re a commodity trader that isn’t using AI-based predictive climate risk signals (not basic weather data), get moving! And, you know, reach out :)