History is a set of lies that people have agreed upon.
Napoleon Bonaparte
Macroeconomic analysis often comes across in the form of nicely formatted charts, illustrating relationships between economic variables such as the labor market, inflation or GDP and the financial market, most importantly interest rates and the stock market.
There is little debate that these links exist and are in fact quite pronounced. However, as Paul Samuelson nicely pointed out once:
The stock market has predicted nine of the past five recessions.
In other words, understanding the lead/lag relationships and correlations between economic data and the financial markets is quite a challenging task and one thing that complicates it further: Revisions.
The publication of economic data always involves a trade-off between timeliness and accuracy. To provide stakeholders with information on the state of the economy as quick as possible, government agencies release indicators before they have gathered all relevant input data. Once new data points are known, the initially released data gets revised, a process which often takes years until a final estimate is reached.
This genuinely sensible process creates pitfalls for researchers analyzing the data at a later point in time. Most importantly, it can induce severe look ahead bias.
Many databases show only the most recently published time series which puts investors at risk of overlooking vintage data and disregarding the magnitude of subsequent revisions.
Keeping this problem in mind is particularly important in the context of studies of turning points. Eventually, we would all love the predict the peak of the boom and the trough of the crisis and economic data can be a bit of a siren call for market timers.
The paper illustrates this using some selective examples. The Economic Revisions Model comes with some preset indicators but beyond this gives users access to the complete universe of ALFRED (the archival economic database of the St. Louis Fed. Just search for a ticker on their website (link is also given in the app), add it to the flexible selectize widget, load the data and get the stats on selected turning points from the 70s to the Covid crash.
The longest time series available on ALFRED is US industrial production and loading this dataset can take a bit of time if the complete history is requested. Once the dataset is loaded, calculations should run quite fast though and the date pickers allow you to select shorter time frames in the first place.
.jpg)
