For the moment, the current nowcast still represents improvement over the official 1.7% real annualized growth rate for the second quarter. The key question: Will the nowcast continue slipping between now and the scheduled release of the official Q3 estimate next month? The answer, of course, depends on the data updates in the weeks ahead. Meantime, cautious optimism prevails, but there's a lot of data to digest between today and the end of October. If the nowcast falls further in the updates to come, that would be a worrisome signal.
Here's where we stand today: A nowcast for the third quarter that's above the previous published growth rate:
But the combination of updates in the economic and market factors over the past week has pared the outlook.
It's likely that we'll see further declines in the nowcast as we move closer to the official release of the Q3 estimate from the Bureau of Economic Analysis. The economy isn't strong enough to inspire expectations that a substantial upturn in the growth rate. But it's also premature to rule out a Q3 that more or less matches the pace in Q2.
Predicting is risky business, however, and so the main value of nowcasting (computing a series of forecasts as new data arrives) is providing a benchmark on the directional trend of estimates through time. In theory, a robust model will become more accurate as the release date of the data point in question approaches. Why? Because the model will process more timely data that influences the number we're trying to forecast. On the eve of the Q3 GDP report, forecasting that number should be relatively easier vs. forecasting two months in advance.
It's important to keep in mind that every prediction model will suffer the indignity of error. That's simply the cost of doing business when uncertainty about the future dominates. But within that broad caveat lies quite a lot of variation. As for the Capital Spectator's model, here's a brief look at how it stacks up in terms of its in-sample history.
As the chart above shows, the nowcast (the red bars) has provided a reasonable benchmark of the actual GDP data. It's hardly perfect, but nothing is in the precarious realm of forecasting. The next chart looks at the same in-sample prediction record going back to the early 1970s.
Eyeballing the track record can only tell you so much. For a quantitative assessment of the model's record, I analyzed the nowcasting history from several statistical perspectives in the software R. The basic result: the numbers are encouraging. For instance, I compared the model with an ARIMA forecast that's optimized for the GDP data set (to be precise: ARIMA (2, 1, 0) with drift). The Capital Spectator's model suffered a moderately lower root-mean-square error (RMSE) rating vs. the ARIMA model (40.2 vs. 59.3), which means that the nowcast errors were lower by about one-third. (The errors here are defined as the dollar amount of deviation, in billions, from the actual GDP data.) I also compared the model against a naïve forecast, which is simply assuming that the previous quarter's GDP will repeat in the next quarter. Not surprisingly, the naïve prediction's RMSE was higher still: 87.1
Although the GDP nowcasting model looks relatively robust on these and other tests in terms of its in-sample history, the true test is how it performs on an out-of-sample basis. In other words, the model's performance in the run-up to the next release of GDP data is the only test that matters. Perfect accuracy, of course, isn't possible. Instead, I'm looking for a benchmark that, through time, offers guidance on thinking about the next GDP report. In particular, I'll be closely watching the evolution of the nowcasts between now and the official publication of the Q3 GDP report. Are the updates rising, falling, or holding steady? Stay tuned….
^ Although real personal consumption expenditures is one of the nowcast model factors, real retail sales is used as a proxy until updated PCE data is published.