Visualizing revisions tracking GDP with our Point-in-Time data

Amid daily news surprises from Trump's Washington, we can expect more volatility in economic data. This week, we're visualizing US gross domestic product revisions to demonstrate the use case for CEIC's market-leading Point-in-Time (PiT) datasets.

Feb. 27 figures from the Bureau of Economic Analysis showed that fourth-quarter GDP rose at an annualized rate of about 2.35%. That marked an upward revision of almost 0.1% from the previous estimate. released on Jan. 30.

Our PiT datasets aim to manage the potential complexity stemming from these revisions, which can pose challenges to modeling and backtesting. Even such a slight revision can result in substantially different forecasting outcomes.

The BEA releases three "vintages" for a given quarter's GDP. The "advance" or "flash" estimate is released about a month after the quarter ends. The second estimate follows a month later, and third (or "final") estimate is released a month after that. (A separate round of annual updates can sometimes revise that last five years of "final" revisions.)

Visualizing this process for Q3, we can see the 2.83% growth pace reported in the Oct. 30 release was followed by only a tiny revision on Nov. 27.

But on Dec. 19, there was a strong upward revision to 3.07%. This had the equivalent of turning 3Q GDP from a "miss" into a quarter that surpassed expectations.

This had significant consequences in the markets. As our second chart shows, the dollar broadly weakened after the October release - and jumped after the December revision.

Our final chart shows the consequences of revisions on forecasting. Consider a scenario where you're testing how a GDP forecasting model might have performed before that 3Q release. Using the July BEA report - rather than the revisions released two months later - resulted in a different GDP forecast trajectory for our model.

If models are trained on revised data, they can be distorted by artificially incorporating information that wouldn't have been available to forecasters at the time. This is known as "look-ahead bias." Our PiT offering is an "archive" of that economic data as it was originally reported - and includes more than 10 years of meticulously preserved revisions on data for G20 and ASEAN-6 economies.

If you are a CEIC user, access the story here.