Indicio and Macrobond beat the street predicting inflation and US housing

Our partnership with Indicio Technologies has once again shown how Macrobond users can get ahead of the curve

Over the past week, we first used the machine learning platform to generate a “nowcast” for US personal consumption expenditures (PCE). ‍

Then, we nowcasted the S&P/Case-Shiller index of US housing prices. ‍

The results?

The Case-Shiller report showed March house prices in the largest 20 US metro areas fell 1.15 percent year on year. The consensus estimate had called for a 1.6 percent decline – but Indicio did better, forecasting a 1.29 percent retreat.

As for PCE, the actual release came in in at 4.36 percent for April, considerably higher than the 3.9 percent consensus estimate for this measure of inflation.

Our Indicio model was much closer to the mark, predicting a 4.26 percent gain, as the following visualisation shows.

How did we get this result – and how does Indicio work?

First, some background.

“Traditional” macroeconomic data is released quarterly or monthly at best, and we are only able to observe real-life figures with a significant time lag. On top of that, this data is frequently revised. Therefore, forecasting models rely on univariate and multivariate regression techniques that crunch higher-frequency indicators in weekly or even daily series.

Indicio lets Macrobond users easily work with these univariate and multivariate time-series models while incorporating the wealth of data available from their Macrobond subscription. Combining and weighting these models based on their historic accuracy, the user can potentially create a super-forecast that can outperform any single model.

In the case of PCE, Indicio helped us select several monthly and weekly variables by measuring their statistical influence. We then used these variables to generate different univariate and multivariate models, including various vector auto regression (VAR) and vector error correction models (VECM).

We retained the following indicators: U.S. CPI, retail trade, U.S. core PPI and retail gasoline prices.

As a final step, Indicio allowed us to generate a single forecast, weighting the various models estimated based on their stepwise root mean square error (RMSE).

April values for our four explanatory variables had already been released. Using these, we generated a scenario forecast in Indicio – effectively creating a nowcast of PCE.(This chart also shows the estimates from the various relevant VAR / VECM univariate and multivariate models.)

On to the S&P/Case-Shiller index. With global real estate in focus as interest rates rise, we used Indicio to generate a forecast for one of the best-known measures of US house prices.

We opted to keep 24 multivariate models, using stepwise RMSE (a measure of historic accuracy) to weight each input model. Here’s a chart of Case-Shiller’s trajectory, the bearish forecast Indicio generated and its confidence intervals:

Here is a visualisation of the prediction and reality after Case-Shiller was released on May 30. The actual Case-Shiller figures are in dark blue. The “street” analyst consensus is in purple. And the Macrobond-Indicio nowcasts are in green. Indicio’s prediction of a 1.28 percent decline was closer to the 1.19 percent reality than the 1.6 percent consensus.

We’ve also added a nowcast for April, along with confidence intervals. Our Indicio model predicts the US housing slump will deepen.