A very common statistic that is used without much thought is R². A person with knowledge of basic statistics and mathematics will become nervous when they see institutions make important decisions using R². In this article, I will try to shed some light on both R² and hedging.
Definition
R² by definition means coefficient of determination, and it indicates how well a model fits a set of data. A consumer of commodities typically is exposed to price fluctuations of raw materials. A few typical examples are an oil refiner exposed to crude oil prices, a soyabean crusher exposed to soyabean prices, etc. One of the key factors in deciding whether to hedge oil or soyabeans resides on the fact that if the hedge is good enough to serve the purpose. The term “good enough” is relative but very important.
For example, according to the US accounting laws an 80 per cent R² is required to decide if hedging is acceptable. Hedge accounts cannot be passed through the balance sheet directly unless there is R² of 80 per cent or above. If R² is less than 80 per cent, then hedge accounts are pushed through the profit and loss statement. This causes volatility, which in turn leads to negative sentiment both in the market externally (volatile share prices), and internally among the management.
Fear of failure
Similarly, government-owned entities shy away from the idea of hedging because any loss will be construed as management failure. They thus look for high correlation in the hedges and despite that, the institutions do not achieve their goal. In any case, an improper understanding and use of R² lets people down. So what are we missing? First, R² does not paint the whole picture as it merely states if the model fit is good without addressing the reasons.
Secondly, it is important to see how the plot of errors looks like. For example a linear model might suggest a R² of 89 per cent and a cubic polynomial model might suggest a R² of 80 per cent. The scatter plot for the errors (difference between actual versus predicted values) could actually suggest the polynomial model is better suited for hedging despite a lower R². Third, do not be happy if the R² is 98 per cent as it might suggest over-fitting. In general, over-fitting situations offer little predictive capacity.
The onus is on the supplier of hedging instruments (banks, dealers, advisors) to innovate products that suit customer needs. Selling an ineffective product with a high R² or rejecting an effective hedging tool with a low R² is not wise. The products currently available in the market such as swaps, options, accumulators, digitals etc. are all standard products that only sometimes cater to the needs of the customer. A good understanding of R² helps in innovation of new products, creates demand, and reduces costs of hedging. The cookie cutter approach in commodity hedging is passé, and smart institutions are already looking to insightful solutions elsewhere from data science firms.