This write-up tries to address the question of how we can increase the synergy and collaboration between quantitative and fundamental aspects of commodity investing. As a first step we need to define what we mean by quantitative and fundamental commodities research or investing and what we hope to achieve with each of these methods.
A clash of quant and fundamental points of view
Broadly, in the commodities space, fundamental analysis can be thought of as the study of the balance sheets of different commodities of the major producing an consuming nations. A fundamental analyst tries to forecast the balance sheet given some assumption on the prevailing economic climate, the weather, trade flows as well as other data. After the forecast is completed it is compared to the most recent balance sheet. If the forecasted ending stock are greater (lower) than those reported by the latest surveys it gives a bearish (bullish) signal. Fundamental analysts can also use this technique to gauge the risk associated with the components making up the balance sheets of the different commodities. This process is sometimes referred to scenaria analysis. Here an analysts can study how historical incidents influenced the subsequent price behaviour. An example would be a prolonged drought or a proliferation of disease amoungst a herd of swine. In this way a fundamental analyst is interested in change, or how an underlying fundamental can change and the possible influence it can have on the price of a particualar commodity or relative value pair. This view is very narrow and specific. From a fundamental portfolio point if view this amounts to concentrated positions with well defined and understood risk factors.
The quantitative approach, on the other hand, looks to find repeatable patterns accross a collection of commodities, time-frames or instances. This approach requires a much larger amount of suitable data to test a hypothesis. From a quantitative portfolio point of view this amounts to entering multiple small bets accross a large collection of instruments and managing the risk efficiently, i.e. cutting losses and letting the winners run. The method is highly relient on diversification and the law of large numbers.
Fundamental analysis relies on change and quantitative analysis relies on repeatable patterns. Notice that the two methods are looking for exactly the opposite insight and manage portfolios in completely different ways.
A typical conversation between a quantitative and fundamental researcher can be frustrating for both sides:
- While the quant is waiting to hear insights that are akin to equations, something that can be programmed and backtested on the past data, the fundamental analyst is trying to explain that he looks for unique aspects that are about to change which is unappreciated by the market.
- While the fundamental analyst is waiting to see a list of out- or underperforming commodities or ralative value pairs, the quant is trying to explain that the hit rate of any quant model is slightly over 50%, making individual commodity picking from that out- or underperforming list a coin-toss.
Among many possible ways to improve quantamental investing the main pillar is the power of great questions. Questions take the conversation one level higher, out of the what and how to measure to the why measure anything in the first place. At the level of great questions, quants and fundamental analysts can have a conversation full of insights. Then each can go back to their methods and figure out the what and the how to measure in order to answer those questions.
In the following we state a couple of ways that fundamental reseach questions can be tackled.
In practice a simple technique to quickly test the possible validity of fundamental or other data driven question is with the use of an event study. An event study uses a collection of histical price data in conjunction with the fundamental or other data you want to investigate. Note that the data needs to have time stamps in order to perform the analysis. We then define events as the crossing of certain thresholds of the data we want to test. The method then determines the statistics of the ensuing price dynamics after the event has been triggered. Below we show some examples of event studies we have performed in the past
- Net Managed Money Event Study
- Calendar Closing Stock Model Event Study
- Bridgeton Event Study
- Corn Planting Progress vs December Prices Event Study
Naturally these results can be augmented with a quick backtest, but poor results on a event study will also lead to poor results on a backtest.
We have been using the method of extracting the feature importance from ensemble learning methods to identify the key fendamental drivers in prices, relative value and calendar spreads. If the fundamental researcher has a collection of different features that he would like to test for predictive qualities this is a great method to use. In practice we have used this method in the following write-ups
Another way we use the collaboration between quantitative techniques and fundamental analysis is with the forecasting of fundamentals or items that might in turn influence the underlying fundamentals. Here examples include
- Hog Capacity Constraints
- Chicago and Matif Wheat Correlations
- Maize Parity Correlations
- Calendar Spread Seasonal Entries and Exits
Naturally these ideas can be extended and improved upon. It might even be worthwhile to test how an oracle model with perfect foresight would have performed in guessing the different fundamentals.
What happens to those reseach projects that do not turn out the way you envisioned? The concept of a research graveyard is important and serves as a quick reference to remind us about past attempts that have failed. This is great because it will stop you from redoing the work or trying to reinvent the wheel. In some cases it might even remind you not to keep on beating a dead horse.