In the midst of February’s volatility most of the reasons given to explain equity market declines centred on the role of rising US bond yields and, perhaps, the exaggerating role played by exchange-traded volatility products.
Since then, equities have seen another (modest) bout of weakness (as shown in the red circle below), but against a background of stable US 10-year Treasury yields and an apparent abatement in inflation fears.
To explain such moves, commentators have had to seek a different narrative. In the main, they have settled on signs of weakening macro data (albeit with some mention of the potential impact of trade wars in the wake of US tariffs).
Such short-term price action is usually little more than noise and hardly worth commenting on. However, one interesting feature of attempts to explain the moves has been the use of the Citigroup Economic surprise indices.
Surprise indices look at macro-economic data outcomes relative to expectations and are useful for a couple of reasons. First, they provide a means of looking at a whole host of data at once. Second, and most important, they acknowledge that it is often the delivery of facts relative to expectations rather than the levels themselves that matters for markets.
And these indices have indeed turned down in several economies, most notably Japan and Europe which are currently negative, indicating that expectations are being disappointed.
However, while there is information in these measures, we should be wary in overstating their importance. This is always true of any single variable but there are also a couple of features specific to surprise indices that need to be taken into account.
The index was never designed for this purpose
Citigroup Economic Surprise indices (CESIs) were originally designed to provide trading signals for currency moves over the very short term (originally over a time horizon of just one minute). We have used the indices to explain currency spot changes over longer periods (most notably here and here in posts which still appear relevant given recent US Dollar moves).
Accordingly, the various data points that make up the index (such as jobs numbers or industrial production) were weighted according to how much surprises in each data point had previously (largely in the period since 1998) impacted currency moves over such short time horizons.
Unfortunately the macro variables that investors will pay attention to will change over time: the US trade balance used to be the most important number for markets, today most are on the lookout for signs of wage inflation. Indeed, there are significant periods when investors don’t pay much attention to macroeconomic data at all (for example in periods of geopolitical stress or when corporate dynamics are unrelated to the broader economy).
Investors’ shifting attention is the very reason that price impact was used to establish the index weightings, and why these weights change over time (they are reviewed on an annual basis). For example, from this month onwards the US CESI will no longer include the unemployment rate, or leading indicators, in the suite of 32 data releases included in calculating the index.
Now, there are very good reasons why the variables that currency markets are interested in over the very short term are the same that matter for other assets, but this will not always be the case. It is perhaps unsurprising therefore that CESIs have been found to have little relationship to stock price moves. They were never meant to.
Inherent mean reverting tendency: in index construction and in human forecasting
Another feature of CESIs that make them less appropriate for drawing conclusions about the prospects for many assets over longer time horizons is their natural tendency to mean revert. This is a deliberate feature of the index: genuine ‘surprises’ should be equally likely to be positive as negative. This means that in the long run one should expect the average economic surprise to be zero.
CESIs measure how data from the past three months compare to median expectations. They also employ a decay factor, so that a big surprise today will influence the level of the index more than the same big surprise three months ago. This feature alone would suggest that extreme levels in a CESI are almost inevitably to be followed by declines and we can see this in the longer term history of the series.
We can add to this a behavioural categorisation of how economists make forecasts. Human beings update our default beliefs in the face of new information; if an economist believing that we are in a low inflation world is repeatedly surprised by higher inflation data we should ultimately expect her or his inflation expectations to be revised up.
Behavioural biases could accentuate the mean reverting tendencies implicit in the index. Anchoring can mean slow reactions to initial surprise, while after a prolonged period of surprise there is a tendency to extrapolate trends as we update beliefs. Herding effects would only magnify these forces at an aggregate level.
It is absolutely correct that fundamental developments need to be considered relative to investors’ prior expectations if we are to assess their implications for asset prices. However, a number of issues in the construction of CESIs suggest that they will not always capture the most relevant forces in this respect. It is also the case that these indices in isolation do not capture the difference between data that is generally improving, but less than expected, and data that is deteriorating even more than anticipated. Ultimately if macro and profits data are very disappointing we can expect equity prices to fall, but in such cases the CESIs are unlikely to be telling us anything we don’t already know.
In May last year, Citigroup themselves cautioned against the misuse of CESIs. However, like the Sharpe ratio (and volatility itself) there is a huge temptation to take a metric with some use and overstate its importance.