What does Riskalyze Risk Number really measure? Disclosure of calculation and methodology essential for meaningful discussion.
Riskalyze has responded to our criticism of the math behind their interest rate stress test. Unfortunately, the response does not address a single point of substance and seems rather personal. Our systems serve some of the largest asset managers in the world and top advisors in the US. We have published articles in scientific journals such as Journal of Risk Management in Financial Institutions, Journal of Asset Management and Journal of Risk Model Validation. In the world of scientific risk management, where we come from, it is completely acceptable to question math of competitors and to receive substantive answers. Let’s keep the discussion mature and stick to the facts. Our original advisorperspectives.com piece was removed because the management of that publication did not want to be part of the continuing argument. They have confirmed to us that they do not have any opinion on the validity of the points that we made. But information wants to be free , so you can read it here and be the judge.
Summarizing our original article, here are few simple questions.
Does Riskalyze ‘Risk Number’ really measure risk?
The article states that a ‘Risk Number’ goes from 45 to 65 in an interest rate stress test. That is an increase of ~42%. The article then states that risk goes up 70%. Did the risk go up 42% or 70%? How do we explain this inconsistency? If a Risk Number does not accurately measure risk and should be replaced by some other measure, as the Riskalyze author does in this interpretation of stress test, then what does it do? What is the science behind these results? We could not find a single paper in an academic publication and would appreciate a reference.
How can you say that interest rate risk of portfolio with duration of 6.23 and a 30% stock position goes up “dramatically” in a 1% increase?
Interest rate stress tests always and everywhere work with a measure called duration. Finance courses in college start with duration and most sophisticated traders use the same duration concept. Riskalyze interest rate stress test does not even mention duration. Their original article comes to the conclusion that “…that a 1% rise in interest rates has such a dramatic impact on downside risk is often eye-opening.” But we are talking about a portfolio with duration of 6.23. So, if we only talked about the bond position, it would lose 6.23% (and if we account for convexity as we should, it would lose less, because convexity helps whether rates fall or rise). Of course, 6.23% loss doesn’t sound as dramatic as the dubious 70% increase, but it is the simplest and most accurate answer.
In addition to 6.23 duration, the portfolio in question contains a 38% stock position (Vanguard Total Stock ETF plus Apple), which will likely go up when rates do (rates and stocks have a fairly strong positive correlation). How can you ignore the correlation between interest rates and stocks when you are performing stress tests? Has this methodology been backtested in any scientific publications?
We have done risk modeling for our whole professional careers. Questioning each other’s math in a public forum is absolutely normal. It benefits the public and even the providers as it forces them to be rigorous. That is, of course, if the discussion is substantive and mature.