Friday, May 1st, 2020

No Crystal Ball

Without a crystal ball, we cannot know whether there will be a second and deadlier wave of the pandemic later this year, or whether this crisis has left us more vulnerable to one of the major geo-political or financial risks that we thought we had contained. It may be that we recover from the virus only to realise that too much damage has been done to the global economy to allow any hope of a quick recovery, that earnings multiples for 2022 are too high, and that the equity market will have to refinance the balance sheet of the entire corporate sector.

There is no point in trying to forecast all of this. It is much better to rely on the stock market’s role as an indicator of sentiment – a forward-looking aggregation of future risk and return, with its own internal and sometimes mistaken logic. Our probability models have been doing this successfully for the last 15 years of live-running. We outperform on a risk-adjusted basis and in absolute terms, and are still doing so. However, the speed of recent moves has led us to introduce a new version of the model, which looks at daily – not weekly data. The reason is that central banks round the world are far more proactive than they were, when we started. Apart from the change in time-period, everything else is the same.

As of 30th April, the US equity / bond model suggests that we ought to have an 65% weight in US equities with 35% in US 7-10 year Treasuries. The simple model does not hold cash and is not allowed to play the yield curve. All we want is something which will tell us what to do with our exposure to equity risk in real-time. It is a trading view; it is absolutely not a strategic asset allocation tool. The question is whether we should be adding or reducing; so, we place as much emphasis on whether the indicator is rising, falling or trending sideways and whether is above or below its short-term moving average (MAV).

Here, we run into a problem because if we show output as a simple probability, we get a long series of zero-percent weights in equities, followed by a gradual improvement from zero to 5% and then a sudden acceleration as we get through a level around 10%. It is potentially misleading to put an MAV through this data, if they are constrained by a zero-boundary. However, we can change the way we display the results by recasting the output as the difference between the current run-rate of equity returns and the risk-efficient hurdle rate, all divided by the realised volatility of the equity index: (run-rate minus hurdle) / equity volatility. The underlying calculations are exactly the same, but the display is different. Everything is measured in standard deviations, not percentage probabilities.

If we look at the evolution of this signal since the beginning of this year, we find that it was at a high level in early January but that it fell decisively below its 20-day MAV on 24th January, when the S&P 500 closed at 3,295, just short of its eventual high of 3,386 on February 19th. The message was that investors should stop buying, even if they were happy to hold. On 21st February, when the S&P closed at 3,338. the indicator fell decisively below zero and was also significantly below its MAV, which was the signal to start selling. By the time the index fell below 3,000, on 6th March, the signal was at -2.90 and 1.49 below its MAV. The primary signal remained negative until 29th April, but it moved decisively above its MAV on 24th March, when the index closed at 2,447. At the very least, this was the signal to stop selling, even if investors were too shell-shocked to start buying.

So, where do we go from here? There are three similar episodes we can look at – the summer of 2008, between the collapse of Bear Stearns and the onset of the Lehman crisis, the early part of 2002, after 9/11 but before the US decided to invade Iraq, and the summer of 1998 as the Russian debt crisis morphed into the LTCM crisis.

In 2008, the primary signal was deeply negative and below its MAV in mid-March, just before JP Morgan rescued Bear Stearns. On 1st April, it decisively overtook its MAV and then rose to a peak of about 1.50 in mid-May. The primary signal dropped below zero in early June and remained negative for the rest of the year. It was above its MAV between July 22nd and September 4th. An investor who followed this signal would have bought at 1,351 and sold at 1,236 – a loss of 9% – but would have been out of equities long before the major collapse, which started on September 29th.

In 2002, the primary signal peaked at 1.14 in late March. The narrative at the time was that the Fed and OPEC had combined to flood the world with liquidity and oil and that the US was coming out of recession. It dropped below zero on 12th April 2002, having been well below its MAV for the previous five days, as it become apparent that the US was going to retaliate against Iraq. Though there were rallies, the signal remained below zero through till October, when there were hopes – subsequently disappointed – that diplomatic efforts could avoid conflict.

In 1998, the primary signal rose to 1.63 in late July, as investors hoped that the Asian debt crisis and the Russian devaluation were behind them. As it became apparent that the collapse of Long Term Capital Management was unavoidable, the signal fell below zero on 31st July and bottomed at -3.02 in early September, as Alan Greenspan cut interest rates dramatically.

Conclusions: The average of these three episodes suggests that the primary signal will hit a local peak at about 1.40, compared with the current reading of 0.38 (30th April 2020). At the current rate of progress, it will take another 10 trading days to get there. If the index is going to get back to its previous high, the peak in the signal will have to be higher and later. Based on these numbers, our view is that there will be a period of significant resistance and tactical profit-taking as the index approaches this level. We do not think that it will regain its previous high at the first attempt.

What happens next is unforecastable, but we will be able to observe and react in real-time., An investor who had followed the equity weighting recommended by the primary signal would have outperformed a 50/50 control portfolio by an average of 10.8% over the three periods covered by this analysis. The average volatility was 420bps lower and the drawdown in each case was about half of that suffered by the control portfolio. In the current crisis, the model has outperformed by 5.4% so far, for volatility that is 750 bps lower and a drawdown which is less than a third of that suffered by the benchmark.

Daily models are available for the other major markets: Japan; UK; Eurozone and China. Apart from China, the recent signals have been very like those generated by the US model. The outperformance statistics are also similar.

Leave a Reply

Full Blog Archive:


Other Recent Posts