Risk management is an important function in institutions having a direct exposure to commodity markets.

Numerous tools have been developed so far but time and again we see that these tools fail to capture the tail event that the institution cared about the most.

There could be many reasons why such events are not captured by the existing risk management frameworks in the most sophisticated institutions.

In my opinion, there could be a couple of reasons.

First, people’s misinterpretation of risk gives birth to irrational risk management tools. Second, risk management tools are outdated causing end-users of these tools to make irrational decisions. It is time that risk management, technology and markets are aligned to enable faster and efficient decision-making.

Consider an example of an asset such as soyameal which in the last week has troubled more people than one would have expected. There were many reasons for the sudden rally but the key question is: were the models existent to pick up the cues from the market and dynamically adjust the risk?

Traditional models are mostly statistical, and this has a fair number of limitations. These statistical models are used primarily because of established technology and understanding by risk managers.

Risk managers in banks and trade houses, in my opinion, would little understand any other methodology giving rise to a limitation in the measurement of risk.

Thus, traditional statistical models would have failed to capture the risk in the volatile market witnessed last week.

In today’s markets, application of models based on neural networks might be more appropriate, especially, in this particular case of soyameal.

Neural networks are useful when there is high degree of complexity in behaviour, unknown behaviour between the independent and dependent variables, and a lot of noise in the data.

In addition, when this is combined with the technological prowess of the data science tools available, the risk management can be more effective.

The key question is: do neural networks or statistical models offer the best predictive power in soyameal risk measurement?

In my opinion, the neural networks will offer the better solution but do not take it for granted as one can easily perform a test to compare two models on the same data set to come to a conclusion.

What set me thinking on this course was that a fair number of agriculture players were short on liquid soyameal contract against non-liquid protein inventories.

A sudden rally in the market caused the risk to dislocate, and most of these traditional players used statistical models to capture such risk that failed.

To blame the market is not a solution, to find a risk management tool to adapt to the market is definitely a desired solution.

The writer is based in London and is the founder and Managing Director of OpalCrest. Views are personal.

comment COMMENT NOW