Itaú BBA - Is the Correlation Between Industrial Production and Confidence Still Valid?

Macro Vision

< Back

Is the Correlation Between Industrial Production and Confidence Still Valid?

May 15, 2017

We show a model that explains this apparent decoupling.

  • Confidence indicators have been on the rise since late 2015, while gross industrial production metrics have been stagnant amid considerable volatility.
  • We show a model that explains this apparent decoupling, recognizing evidence that both the level and change in confidence are important in forecasting changes in industrial output.
  • Hence, although the recent increase in confidence is relevant, the rebound will only gain traction once these indicators approach a neutral level.

Improved Confidence Since Late 2015; Stagnant Production Indicators

FGV/IBRE publishes confidence indexes as a summary of the many monthly surveys on economic trends. In particular, the seasonally-adjusted monthly reading of confidence among industrial entrepreneurs is both a leading and coincident indicator of manufacturing output, averaging 100 between 2001 and 2015.

All confidence metrics improved in 2016, after widespread declines in 2014 and 2015. Industrial confidence climbed to 91.2 in April 2017, rebounding to 2Q14 levels, from 76.3 in December 2016. Meanwhile, confidence rose to 89.1 from 68.6 among retail entrepreneurs and to 84.2 from 68.5 among service businesses.

However, the recovery in confidence was not followed by a recovery in related output data. An assessment of industrial production (related to industrial confidence), broad retail sales (retail confidence) and real service revenues (service confidence) has shown stagnation and volatility since 1Q16.

Given the apparent decoupling between the surveys and gross figures, we wondered if the correlation exists (and persists) or if it began a structural breakdown in 2016. Our focus is on industrial production, given the longer period of the survey sample.

Both the Level and Change in Confidence are Important in Forecasting Industrial Output Growth

We initially built two quarterly models in an attempt to explain the change in industrial production. The first (“change” model) uses changes in confidence as the independent variable; the second (“level” model) uses confidence levels. Both show a significant statistical relationship with changes in industrial output.

We designed a back-test starting in 1Q16 to evaluate the models (chart below on the left). The two models show systematic errors since confidence began to rise. The level-based model underestimates production growth since 2016, while the change-based model overestimates growth during the period

Given the two previous models’ shortcomings in forecasting recent developments, we designed a third model combining the change in confidence and its lagged level.

The surveys’ methodology is consistent with this mixed relationship (of both level and change). Some of the questions that survey participants answered to create the indexes mention change (allowed answers are more, less or same) and some mention levels (such as strong, regular or weak).

These two factors show a statistically significant relationship with industrial production (see table 1). Furthermore, the models do not rule out that 100 is the “neutral” level throughout the sample.

Applying the same back-test used in the previous models, its results approximated the actual industrial production readings (see chart above on the right), without the systematic errors of the two initial models.

Assessment of retail and service confidence surveys is more difficult because samples are shorter and have not undergone several cycles of expansion and contraction. Results tend to favor confidence levels (over changes in confidence) to explain changes in their respective gross figures, unlike industrial production.

All in all, the models suggest that the relationship between industrial confidence and industrial production remains valid. The rebound in output is delayed because confidence is now significantly below the level regarded as neutral (approximately 100 in the three indexes). Hence, the low confidence level has been offsetting the positive effect of the gradual growth in industrial production.

As confidence levels approach the neutral level, the correlation between the change in this measure and industrial production growth should once again become evident.



Artur Manoel Passos

Alexandre Gomes da Cunha


< Back