Everything's Correlated -- Where we are in this drawdown

There have been energetic discussions on the investment losses of specific asset classes from COVID-19, but less on the loss that a diversified multi-asset portfolio might have suffered so far. I ran a quick calculation using our system to show this here.

With such sharp market moves, seeming price inconsistencies have rippled across different types of investments, and the impact on a diversified portfolio is non-obvious (other than most likely being down). Of course, everyone's portfolio can be very different, but for the purpose of simple analysis, I will take what would be a reasonable, and somewhat conservative, multi-asset portfolio.

Hypothetical Diversified Portfolio

This portfolio is highly liquid and therefore market pricing is accurate, with the credit components being the least liquid during highly stressed markets. Calculating the portfolio simulation, the recent drawdown for it looks like the following in the context of the previous 2 years of daily performance.

Simulated Drawdown

The drawdown started after Wednesday 19 February, and has been shown here until close on Monday 16 March, with a loss of 10.6% (in log terms, or 10.1% in linear terms). There might of course be further drawdown later, as this article was written on the 17 March. Scrolling this window back shows during the global financial crisis the drawdown was 17% (in log terms) and during the dotcom bubble recession, the drawdown was 10%.

Foreseeing the Risk

As much as it might not feel like it, the current drawdown caused by COVID-19 is a known-unknown risk and not an unknown-unknown uncertainty (aka. Knightian uncertainty). One of the issues though comes when we try to gather data to measure this risk. There is almost no relevant data, or at least none with the technological and situational context of today. The 1918 flu pandemic is the most "relevant" event, however hard data is extremely sparse from such a long time ago, and response mechanisms available today are entirely different.

So rather than use sparse data, we use a trick, and in so doing, we expand the data set to include all risk events; Knightians and Black Swans included. This gives us a lot richer data on the one primary concern we have as investors, which is cumulative capital loss (or drawdown).

The trick is to flip the risk around, and ask: Within a defined time period (I generally use 6 years, but can also be e.g. 3 years), what would be the worst drawdown you would expect during this theoretical period? This concept captures all risks and their likelihood of occurring into the overall expected cumulative capital loss in the defined period. The focus is on the immediate future 6 years (as we like to drive looking out of the front window), although we can and want to examine this ex-post as well by "rolling-back" a 6-year window as far back as we can, to help inform our ex-ante views.

Importantly, the risk event is not specified. It could be, among many others, a:

• pandemic (COVID, 2019),
• terrorist event (September 11, 2001)
• bank collapse (Lehman Brothers, 2008)
• Ponzi scheme (Madoff, 2009)
• tsunami (Tōhoku and Fukushima, 2011)
• sovereign bond default (Greece, ~2012)
• or a global solar storm (Carrington Event, 1859, and also the very fortunate near miss of a solar storm in July 2012)

By not specifying the risk event, we allow substituting unknowns for other unknowns, we care that something not good could happen within a 6 year period (even a solar storm, which we have essentially zero data). From an investment perspective, we then care what we should expect to suffer as a drawdown from the unknown event(s), which is called the Expected Drawdown.

Doing this for a single asset is relatively intuitive. Expanding to a multi-asset portfolio context is also intuitive, but requires a mathematical framework and a number called Expected Co-Drawdown. This framework is something I have written about in detail here if you haven't read about it already.

Expected Drawdown Calculated

So coming back to our diversified multi-asset portfolio, what was the portfolio Expected Drawdown before COVID-19 hit? Using FountainArc's automated modelling exclusively (i.e. not yet coloured by personal ex-ante views), here is the portfolio's ex-ante Expected Drawdown, E(DD), and contributing drawdown risks from each of the positions, d(DD).

The portfolio's Expected Drawdown as at January, before the COVID-19 drawdown, was calculated as 12.9% (log basis). This is saying that we should have absolutely expected a drawdown of 12.9% (12.1% in linear terms) with this portfolio at some point within any 6-year period, coming into the COVID-19 event.

Where We Are

First and obvious point is that the (so far) actual drawdown of 10.6% (up to 16 March 2020) is still slightly smaller than the Expected Drawdown of 12.9%. As bad as the recent few weeks have been in the financial markets, we could have expected and quantified something of at least this scale before it happened.

The Expected Drawdown is therefore an intuitive way of providing to all portfolio stakeholders an idea of the scale of risk that is being run, and bringing that risk into expectation ahead of time. Ultimately, the portfolio drawdown risk can be managed directly to the level acceptable, and used as the mechanism with which individual positions are sized.

Can this drawdown get worse? Yes it certainly can; we haven't even (quite) reached expectation. I am often asked about the range around the expectation. This gets into further technical discussion. For those familiar, in a theoretical CoDD1 scenario we are looking at a drawdown in this multi-asset portfolio of 18% (16.5% in linear terms). Modelling a cascading fallout into deep recession using the system's Stress Testing - such as the global financial crisis after Lehman collapse - would be the region of 20% (18% linear).

Therefore in the worst reasonable scenarios, we are slightly over halfway there for this particular portfolio. Historically, these worst type of larger drawdowns have occurred over many months with intermittent partial recoveries.