Sean Carr, Senior Vice President and Global Market Strategy Director contributed to this article.
Enterprise risk management is typically siloed into three areas:
- Market risk, which is related to investment risk
- Credit risk, which is related to obligor default management
- Operational risk, which is the control of bad outcomes due to processes and procedures.
Each silo has developed a toolkit to deal with its unique issues, but the toolkits do not completely address all types of risk neatly. For instance, how should the market impact on reactions to government policy be characterized?
We are talking about policy risk as the adverse impact of changes in governmental policy on investment portfolios. This included changes to regulations, laws, and policy, as well as the outcomes of elections and referendums. The tools of market risk are not well suited to these risks, as they measure the well-observed market reactions to the market itself. Policy risk, by contrast, is episodic in nature.
As such, it is difficult to develop accurate estimates of impact based on repeated historical behavior. Market risk, in essence, is the actions of many individuals in aggregate; policy risk is the actions of a few individuals. Moreover, in most developed countries, policy risk evolves in an open fashion, as government leaders publicly declare initiatives and polls forecast election results. Where policy risk analysis can have maximum utility is when events do not unfold according to expectations. Thus, policy risk comes down to forecasting rare events, which are not already anticipated. Said differently, the goal is not to forecast the probability but the impact.
Forecasting the Unforeseeable
The first step is to develop a forecast of the unexpected policy outcome itself. These forecasts are usually derived from experts or consultants. There are well-known methods on how to utilize and combine expert opinion.
The next step is the bridge between policy outcome and market risk. Somehow, a liquid asset must be selected to be the instrument of immediate and appropriate impact. The asset should also be well correlated to other assets in the markets, as it will serve as the propagation agent.
Since eventually the valuation of financial assets is derived from interest rates, equities, or currencies, we would recommend using equity indices for national events and foreign exchange rates for global events. However, the need is to be extremely parsimonious in modeling with only very few factors. Analysts should avoid expressions of over-confidence in building detailed stress tests and the naivety in the assumptions that can be reflected through these correlations.
Finally, how much will the market move over what horizon? In order to permit comparison we would recommend thinking in terms of standard deviations. What follows is a commentary on the standard approach to sizing of shocks, rather than an in-depth financial analysis.
Analysis in Action
Assume that policy events are what drives the non-normal fat tails so common in financial time series. Shown below are the daily returns, z-scored, and the frequency at different multiples of standard deviation. Pick one you think is representative of the type of impact. Next, given that multiple, what is the expected tail loss? Finally, how long are markets disrupted?
As for the horizon, we contend that short horizons are more appropriate. Once the near term impact has lessened below extreme standard deviation multiples, any longer horizon is dominated by normal market trajectories, and thus outside the scope of this analysis. The table below demonstrates this analysis for equities, interest rates, and currencies, and for different z-score multiples. Finally, we want to reiterate to keep it simple.
The above table seems to suggest that a “generic” stress test should be between four and five standard deviations over a horizon of two to four days. However, as demonstrated below, this should be considered a minimum amount of the dislocation. Consistent with the tone of this note, the true magnitude is potentially unpredictable, and if you’re not comfortable at five standard deviations, you won’t be comfortable at eight or nine. Second, test both up and down shocks. The ultimate market reaction doesn’t have to conform to your expectations.
Considering the recent triggering of Article 50 by the UK to begin the formal process of Brexit, the impacts of the vote last year offers a recent and pronounced policy shift that allows us to test this methodology. The results below show several things: the validity in the selected propagators as they immediately reflected large movements in market sentiment locally and globally across the major capital markets; the challenge of scaling modeled shocks as this result had a much large impact we might have thought possible.
Looking at daily returns for 2016 (roughly six months either side of the event), we can see that while the impact on equity markets was marked (down 133 points, -3.7 STD), when also compounded with the large drop in USD/GBP exchange rate it was almost a nine standard deviation event from a dollar-investor perspective.
To link this to the real world we considered 28 various funds with various asset classes and geographical exposures and ran projected returns to compare with actual returns. Creating the stress tests under different levels of shocks (two, three, and five standard deviations) and also using the measured standard deviations from the Brexit day vote itself, we can see just what a surprise the result was.
In conclusion, given that it falls outside the time-tested and well-observed patterns of market, credit, and operational risk, policy risk represents another complex factor for investment portfolios. Using the method above offers an opening for how we might start modeling such risks, but our example highlights the potential for the real world to deliver shocks far beyond "the norm."