The Sunday Times 28 January 2018
On Sept. 10, 2009 former trader and bestselling author Nassim Taleb testified to the House Committee on Science and Technology. It was a hearing on the responsibility of the mathematical model called Value at Risk (VaR) and its role in the financial crisis.
Taleb had been a very vocal and long time critic of the use of VaR in the finance industry. He has been unequivocal about the threat posed to the financial system by financial models particulary VaR. Post-2008 financial crisis, his warnings seemed all too prescient. Now that the damage wrought by VaR seemed inescapably obvious prompting an investigation in to its role in the crisis one might reasonably have thought it would have been the end of VaR modelling.
Quite the opposite in fact. The newly introduced PRIIPS regulation (packaged retail and insurance-based investment products), which came into force in Europe at the start of the year, requires providers of PRIIPs to produce a KID (key information document). Forgive the use of all the acronyms, but the financial services industry is in thrall to them.
And the key concept of risk in these new documents? Yes, is based upon VaR (specifically, value at risk equivalent volatility or VEV).
VaR was popularised in the early 1990s by a handful of scientists and mathematicians (known as ‘quants’ in the business) who went to work for investment banks on Wall Street. Its great appeal, particularly to non-quants, is that it expresses risk as a single dollar/euro figure (without the requirement to even bother ourselves with percentages). It is a measure of possible loss on a bad day usually with a 1 per cent probability - so you should expect one every five months, based upon 20 trading days in a month.
Taleb feared VaR was doomed to being a very wrong estimate, because of its analytical foundations (it uses historical data) and the realities of real-life markets (where extreme events occur with far greater frequency than financial models predict).
Worse still, it can be gamed. You can get a VaR that´s very low – all you need is a portfolio of assets that happened to have recently enjoyed benevolent calm and/or little correlation with each other. The VaR estimate will tell you there is little to concern yourself with.
Adjustments to the VaR model to include extreme events are dangerously naïve in Taleb’s view. There is a link between the harm of the events and their unpredictability, “nothing predictable can be truly harmful and nothing truly harmful can be predictable”.
There are low probability events that are predictable. Financial Times columnist John Kay uses the analogy of the tail-gaiting driver that saves minutes on their journey each day, until one day they don’t come home. He says that “tailgating” strategies return regular small profits with a low probability of substantial loss. You can’t predict when a tailgating motorist will crash, but a crash is one day likely..
At the other end of the spectrum are low-probability events to which you cannot attach a probability because you cannot even conceive the event – Taleb’s so-called “black swans”. We can’t assign a probability that someone will invent the iPhone, because to conceive of such a probability is to have invented it.
The financial crisis of 2008 was arguably predictable, but may as well have been a “black swan” for all the utility VaR models provided.
As a measure of this you may recall an interview with David Viniar, then chief financial officer of Goldman Sachs, just as the global financial crisis began to break in 2007. He reported in shock that his firm had experienced "25 standard deviation events, several days in a row". To be fair to Goldman, they were not alone in this.
The distribution on which such claims are generally based would estimate the percentage probability of experiencing a 25 standard deviation event as being roughly a decimal point followed by several lines of this column filled with zeros and then a one.
These events simply do not occur – according to VaR models.
What Mr Viniar meant, affording him the benefit of the doubt, is that events had occurred that fell outside the scope of its models. Or in more direct parlance, the models were wrong. As John Kay so eloquently puts it, “to toss a coin and achieve 100 heads in a row is to witness a bent coin, not a miracle.”
Little appears to have changed at Goldman where Mr Viniar’s successor, Harvey Schwartz, only a few years ago when the Swiss franc was unpegged against the Euro, noted Goldman Sachs’ experience as a "20-plus standard deviation" occurrence."
The regulatory disclosure that past performance is no guide to the future, has appeared in financial advertisements for years, and for good reason. The new risk statistics undermine this tenet. The PRIIPs legislation may be extended to an even wider set of investors and assets held in UCITS (undertakings for collective investment in transferable securities). I fear this will not improve investors understanding of risk. The opposite is more likely.
Gary Connolly is Managing Director of iCubed, an investment consulting company providing investment support to financial advisors. He can be contacted at firstname.lastname@example.org or on twitter @gconno1. iCubed Training, Research and Consulting, trading as iCubed, is regulated by the Central Bank of Ireland.