Once is chance, twice bad luck, but three times – “Enemy Action”
Con Keating here expresses his extreme frustration with USS
USS has been at it again. On March 1 2021, they published a paper entitled: “The difference between our expected investment returns and our prudent assumptions”.
The paper states an objective:
“… it is important for anyone with more than a passing interest in the 2020 valuation to appreciate the difference between forecasts for expected investment returns and prudent assumptions.”
Note the switch from the USS specific of the title to the general here, but don’t take that as gospel – the switching back and forward in this document could leave a reader dizzyingly confused.
I will leave the reader to consider the logic and veracity of the paragraphs leading to this:
“Chart 1, below, illustrates that returns even over long horizons do not follow easily predictable patterns and extrapolation from history is fraught with danger. For example, US equity returns experienced over the past 20 years have been substantially lower than what one would have projected on the basis of the experience of the 1980s and 1990s as a result of the impact of the dot-com bubble and the Global Financial Crisis.
Chart 1: Experienced vs. extrapolated US equity returns over rolling 20-year periods
The ‘extrapolation’ here is no more than transposition of the original series twenty years to the right – that is a bizarre conception of extrapolation. Now, I don’t have a series or returns for US equities going back to the early 1800s and frankly, I would doubt that events before 1900 had any meaningful relevance for today. However, I do have to hand the CRSP dataset going back to January 1926 – which has been the workhorse for more academic studies that I have time to count.
From those, I calculated rolling twenty, forty, and sixty-year averages of these returns – the twenty and forty series are shown in figure 2 below. By way of extrapolation, I run linear regressions on each of these series. The equations for these regressions, which are also known as trend lines, are shown on the chart. (for just the 20-year and 40-year to avoid cluttering a diagram) They are to all intents and purposes the same, separated only by a small, fixed amount. The R squared for the forty-year series is over twice that of the twenty-year series, and the volatility of that series is much less. The volatility of the series declines as the length of the averaging increases; the 40-year is 1.4 times as volatile as the 60-year, and the 20 year is three times as volatile as the 60 year. The twenty-year volatility is 2.2 times the forty-year series.
Here the important point is that both the twenty-year and the forty-year series regression lines are upward sloping, as also is the 60-year series, implying higher returns on average in the future, not lower. Something in excess of 10% pa is to be expected on the basis of that extrapolation. It happens that the forty-year is currently a little above its regression value and the twenty-year well below. The sixty year is also substantially below its regression value.
By contrast, had we looked at just the twenty-year series at the end of 1999, we would have seen that was very significantly above its own regression line, which of course should have sounded alarm bells for the immediate market prospects.
Figure 2
For the avoidance of doubt, let me say that I am not advocating linear regression as a forecasting method, though it is one simple method. Let us move on. The document states “Here, we can look to the relentless decline over the past decade in the prospective returns offered by ‘risk-free’ assets like government bonds, to perhaps understand why other assets have risen in value (see: Chart 2).”
This chart is bizarre. The choice of equity index is obscure when the objective is to show a relation equities and index linked gilt yields; a total return index rather than a price index. It is also stretching credulity to relate world equity prices or returns causally to UK gilt yields. This chart compares a yield and a price and both scales are linear – to make any comparison validly the scale for the price series should be logarithmic – and it should be price, not total return.
I have simplified this presentation by assuming uniform changes between the start and end dates. This is shown as figure 2 below.
Figure 2: World equity returns and ILG yields
This presents a very different story. The inverse line shown is the performance to be expected if ILG yields drove world equity prices perfectly. The equity return series shows very little by way of response to falling gilt yields, when presented correctly on a log scale, and most of that will disappear when we consider prices rather than total returns, which are inflated by dividend income reinvested. This prompts a series of questions:
Are equities much too cheap? or
Are index-linked gilts too expensive? or
Are both of these true?
If I were to want to make a point about the relation between gilt yields and equity prices and be plausible, I would use 20-year conventional gilt yields and plot this alongside the log of FTSE 100 prices for the period from, say, January 1990 – and the result of that is shown below as figure 3. But the bad news is that the slope of FTSE 100 price series is just one seventh of that of the gilt yield series (as can be seen from the two regression equations – a 100 basis point decline in 20 year gilt yields is associated with only a 6% increase in FTSE 100 prices – just 150 index points on average over this period.
Figure 3 FTSE 100 prices and 20 year gilt yields, 1990 to present
The USS document continues and states:
“Since the 2008 valuation, we have seen the growth in liabilities outstrip the growth in assets on a variety of measures. This is primarily due to falling bond yields (the returns offered by low-risk investments) over the past decade. This is shown at a high level in chart 3 below, which compares the assets at each valuation date since 31 March 2008 with various measures of liabilities over that time.”
and then produces a series of valuations – all, of course, as calculated by USS. This is proof of nothing other perhaps than the incompetence of USS.
The presentation of the cause of the problem, as they perceive it, is telling, “falling bond yields”, and this is reinforced by the following line defining a bond yield: “the returns offered by low-risk investments”. It is actually the decline in the discount rate which has wreaked havoc, but to USS it seems that discount rates are the same as low-risk bond yields. Incidentally, lower risk investments are not confined to bonds and of course, the falling yields of low-risk bonds would have been relevant only to new investments, not to the stock of assets already held in the fund portfolio.
I have focussed only on the three illustrations in this USS article – that is not to say that this is all that is wrong in the article. They are just egregious examples. I was thinking that incompetence might have been the driver, but as the article has no attributed author, there’s no-one to speak to, so I am left wondering a much darker thought: Does it lack an attributed author because no-one in USS wanted to be associated with such ‘Dezinformatsiya’?
When I ask myself the question: would this paper have helped me if I did not know the difference between “expected investment returns and prudent assumptions” and the answer is a resounding NO.
As it happens I did already know the difference, but then I also know the difference between meteorology and climatology.
Con Keating – March 2020
Postscript;
Deeper probing into the USS website finds that the report may have an author. Thanks to Mike Otsuka for finding this link. To quote Mike’s email to me
USS actually attributes an author, but incompetently does so only on a page of “Views from USS” which links to that post (while failing to identify the author in the post itself):
Con’s substantive allegation is that this is Dezinformatsiya’