Model Apocalypto – Is Model Risk Important?
University of Kent, Canterbury – Kent Business School
Assets in private equity and venture capital strategies have seen significant growth in recent years. In comparison, assets in the hedge fund industry have experienced slowing growth rates. Q2 2021 hedge fund letters, conferences and more Over the six years to the end of 2020, hedge fund assets increased at a compound annual growth rate Read More
September 01, 2015
The number of models and techniques proposed for financial applications seem to increase both in number and complexity. Here we present a critique of the knowledge acquiring and subsequent application process in the relatively new field of Finance. Focusing on two of the most widely used concepts in Finance, volatility and risk measures, it seems that it is possible to generate a wide range of scientific based results under the same information set. In addition, the interplay between the generic finance problems on a theoretical level and the applied solutions involving econometrics/statistics inferential approaches reveals that more fundamental research is needed in order to consolidate theoretical knowledge in finance as well as helping out practitioners in selecting the relevant techniques for their applications. In essence, this article is a plea for more conceptual thinking and less model proliferation.
Model Apocalypto – Introduction
Finance was part of the evolution of humanity for thousands of years. Follow the money and you will understand the course of history. The nascency of mathematics was triggered by the need to solve problems related to money and finance. Modern finance has experienced a meteoric rise in the 1980s, coupled with the introduction of computers on a large scale and also with the liberalization of financial markets. Scientists from many other disciplines like Mathematics, Statistics, Physics, Mechanics, Enginering and Economics, found a new uncharted territory in Modern Finance and they embraced the new “gold” scientific race.
After a sunrise there is a sunset, and exuberance quite often masks lack of full understanding of the complexity of problems that may surface at any moment in time. The series of crises in Finance culminated with the subprime-liquidity crisis that started1 in 2007 that was reminiscent of the financial crash of 1929. Who was to blame and what really happened is still the subject of intensive research, and valuable lessons are to be learned overall.
While everybody is offering an opinion about toxic assets and liquidity measures and trying to design measures of systemic risk impact, not enough attention is paid in my opinion to another source of future problems, that could also reach catastrophic and endemic levels, that is the risk carried by different models, or shortly model risk. What is model risk? Is it important? Can we measure it? These questions are very important for financial markets.
Is model risk important?
There are already manifestations of model risk that led to substantial losses and it may sound cliche to say that this is only the tip of the iceberg. In 1987 Merrill Lynch reported losses of 300 million USD on stripped mortgagebacked securities because of an incorrect pricing model and five years later in 1992 J.P. Morgan lost about 200 million USD in the mortgage-backed securities market because of inadequate modelling of prepayments. Bank of Tokyo/Mitsubishi announced in March 1997 that its New York-subsidiary dealing with derivatives had incurred an $83 million loss because of their internal pricing model overvalued a portfolio of swaps and options on USD interest rates. Dowd (2002) pointed out that the loss was caused by wrongly using a one-factor Black-Derman-Toy (BDT) model to trade swaptions. The model was calibrated to market prices of ATM swaptions but used to trade out-of-the-money (OTM) Bermudan swaptions, which was not appropriate.
With the benefit of hindsight it is known now that pricing OTM swaptions and Bermudan swaptions requires multi-factor models. Also in 1997, NatWest Capital Markets reported a $50 million loss because of a mispriced portfolio of German and U.K. interest rate derivatives on the book of a single derivatives trader in London who fed his own estimates of volatility into a model pricing OTC interest rate options with long maturities. The estimates were high and led to fictitious profits. It is not clear whether the trader simply inflated the volatility estimate or he came up with the estimate that was more “convenient” to him. Elliott (1997) pointed out that these losses were directly linked to model risk. Williams (1999) remarked that model risk was not included in standard risk management software and in 1999 about 5 billion USD losses were caused by model risk.
The recent advances of algorithmic trading add another dimension to model risk. It is difficult to say what exactly is happening and who is to blame in this new type of superfast trading, most of it being opaque and difficult to control. A Deutsche Bank subsidiary in Japan used some smart models to trade electronically that went wild in June 2010, going into an infinite loop and taking out a $183 billion stock position. The thing about computers is that any mistakes are executed now thousand of times faster than before. There is no doubt in my mind that the next big financial crisis will be generated by model risk.
Model risk has been identified previously in all asset classes, see Gibson (2000) and Morini (2011) for interest rate products, Satchell & Christodoulakis (2008) and Rosch & Scheule (2010) for portfolio applications, Satchell & Christodoulakis (2008), Rosch & Scheule (2010) and Morini (2011) for credit products, and Campolongo et al. (2013) for asset backed securities. It has also been recognized in relation to measuring market risk, see Figlewski (2004), Escanciano & Olmo (2010),Boucher et al. (2014), Danielsson et al. (2014b).
The concepts of risk and uncertainty have been intertwined. Uncertainty on the other hand is a recognition of the existence of outcomes unspecified that may still occur and with which we do not have any way of associating a probability. Playing cards or roulette falls in the first category, saying whether there is life on a far away planet is an example of the latter and predicting the next type of fish you will encounter when going deep in the ocean is an example where both risk and uncertainty are combined. Likewise, we can make statements about the possible future value of the share price of Apple but we cannot say very much on the source of the next big crash in financial markets. In other words, the share price of Apple is risky while the source of the next big financial collapse is uncertain. The important distinction between risk and uncertainty goes back to Knight (1921) who pointed out that risk stems from situations where we do not know the outcome of a scenario, but can accurately measure the probability of occurrence. In contrast, uncertainty is associated with scenarios where it is not possible to know all the information required to determine the scenario probabilities a priori.
See full PDF below.