Brazil’s economy (Part 3): Consumption
My last article “Brazil’s economic downturn (Part 2): Employment” presented econometric evidence consistent with 71% of Brazil’s average employment growth between 1998 and 2016 being attributable to Brazil’s internal economic sector, and 29% attributable to the external economic sector; predominately related to Brazil’s foreign exchange rate and related factors. The result is fairly interesting because only about 8% to 16% of Brazil’s economic activity is attributable to imports and exports over the 1998-2016 period. Why does Brazil’s external economic sector appear to have roughly twice the influence on Brazilian employment than would be reasonably expected? And, again, what’s actually causing Brazil’s economic downturn?
In this article I study another dimension of Brazil’s economy that’s closely related to employment: Brazilian private sector consumption spending. I use the same basic approach here as was used to study employment, so I’ll now move a bit more quickly with less explanatory comments.
Econometric analyses of private sector consumption spending and related data from 2005 through 2016 roughly suggest, among other things, that Brazil’s …
(i) consumption growth is attributable to internal economic sector factors, and
(ii) consumption contraction is attributable to external economic sector factors.
Interesting, if true; don’t you think? Please read on …
1. Brazilian private sector consumption, 2005-2016
Private sector consumption spending depends on–i.e., can be funded by–compensated employment, investment income or liquidations, government financial assistance payments, or borrowing (increases in debt). For a variety of reasons including the difficulty of obtaining data on all these factors, my initial analysis of historical trends in consumption will only involve compensated employment and consumer borrowing.
Notice two things about the graph: (i) consumption spending and employment level is highly correlated, although there has been a slight divergence since the beginning of the broad economic downturn in mid-2014; and, (ii) although non-mortgage debt began a downturn in mid-2012, total debt continued to increase until the mid-2014 broad economic downturn. The timing of the non-mortgage debt contraction–or alternatively the continued increase of mortgage lending as non-mortgage debt contracted–being coincident with Brazil’s industrial sector downturn in mid-2012 suggest such debt levels are in some way either causes or effects of the downturn.
Debt levels alone do not necessarily influence the funds available for consumption spending, it is perhaps helpful to also consider a graph for the same period presenting consumption spending and household debt service requirements as a ratio of disposable income (DSR/DI); for both total household debt and non-mortgage debt:
An interesting, reasonable interpretation of the graph is that it suggests consumption spending is generally proportional to–and, so, highly correlated with–the DSR/DI ratios, and that when the relationship substantially deviates from that proportion it tends to return over time to that proportion (e.g., note consumption spending and DSR/DI lines seem to re-converge after diverging).
In any case, both economic intuition and the above graphs suggest that both employment levels and debt levels are important causal factors of consumption spending. I will explore this formally next … .
2. An econometric model of Brazilian private consumption
Econometric model. As with my econometric model of Brazilian employment, it is best for technical reasons to frame the model in terms of logarithmic growth rates (more accurately termed rates of change), which have the basic following form:
Similar to my study of Brazil’s employment growth rates, I tested between 5-10 different models comprised of hypothesized causal factors and found the most accurate, reliable model–in a technical sense–to be the following:
The model was estimated using 2005-2016 quarterly data derived from Banco Central do Brasil time series data. The independent (right-hand side) variables are, in order, …
- Q1: seasonal indicator variable for 1st calendar quarter
- Brazil total employment growth rate
- DSR: debt service requirements / disposal income
- NMDSR: non-mortgage DSR
- INTRATE: Brazilian prime interest rate
- Rate of change in Brazil / US foreign exchange rate
- Rate of change in Brazilian exports
Estimated marginal effects on consumption growth. Conditional on (controlling for) all factors in the model, between 2005 and 2016 the estimated average marginal effects on consumption growth for the factors are as follow (rounded to nearest .01):
- Q1 has a positive marginal effect of .13 (higher in Q1 than other quarters),
- Brazil employment growth has a positive marginal effect of .47,
- DSR has a negative marginal effect of .63,
- NMDSR has a positive marginal effect of .74,
- BRL/USD rate of change has a negative marginal effect of .05, and
- Brazil exports rate of change has a negative marginal effect of .04.
General parameter estimate interpretations. With the exception of Q1, each of the causal factors above is measured as a rate or a proportion. If indeed the relationships are causal, this means the general interpretation is that a .01 (1%) positive change in the factor has either a positive or negative marginal effect equal to to the parameter estimate (which, when multiplied by 100 represents a percentage).
So, for example, if there is a .01 (1%) increase in Brazil’s total employment from the prior quarter, then the estimated average marginal effect of this 1% (.01) change in total employment is a positive .47% (.47).
Inferring causality. With respect to causality, it is always important to consider how the factor causes the change in a fairly precise way. For example, a growth in (compensated) employment causes wages to be earned and received; which, in turn, generally causes people to spend more to improve their living standard through increased economic consumption. The basic idea is that it is generally necessary to specify a logical, empirically demonstrable, step-by-step chain of causes and effects that ultimately link the independent variable(s) with the dependent variable if we are interested in inferring causality via econometric analyses.
I don’t do this explicitly in these articles for the sake of brevity, but would be happy to do so for any interested reader.
Specific parameter estimate interpretations. While I think most of the relationships between consumption growth and the hypothesized causal factors are fairly straightforward, those related to DSR, NMDSR, and export growth rate do not seem obvious. I will propose simple, plausible interpretations here:
— DSR (total debt service requirements as a proportion of disposable income) is negatively related to private consumption growth because as DSR grows there is relatively less funds available for consumption.
— NMDSR (non-mortgage DSR) is positively related to private consumption growth because in the sample period non-mortgage lending apparently funded a significant amount of consumption growth and, when it began contracting in mid-2012, apparently no other funding sources were available to support the existing higher consumption levels; thus leading to a contraction in private consumption.
— Export growth, as well as BRL/USD exchange rates, are negatively related private consumption growth because in any market buyers compete to purchase the products they want, and the low bidder loses the bidding competition and then tends to buy the lower cost less desirable product. Because Brazil is a substantially lower income country than, for example, the USA, Brazilians tend to lose product market bidding competitions to US buyers. Relatedly, when the Brazilian Real devalues relative to the US Dollar, this exacerbates the negative effect to the extent that many global product markets are priced in US Dollars. In short, Brazilian consumption growth decreases when Brazilians lose income and buying power when priced in stronger foreign currencies.
Note that the interpretations may or may not be correct; they simply seem intuitively plausible to me. To actually know if the interpretations were correct or not, it would be necessary to more carefully study these explanations with both more theory and evidence than I’ve presented here.
Econometric model predictive power. The R-squared statistic of .983 of the econometric model suggests that the hypothesized causal factors explain about 99% (!) of the variation in Brazilian private consumption growth. While this is, by any standard, an extremely high R-squared statistic, it is mainly just an artifact of the high degree of seasonality in Brazil’s private consumption patterns as seen in the following graph:
So, the high R-squared statistic is partially attributable to the inclusion of the Q1 seasonality variable, conceptually resulting in fitting the model through the Q1 observations much more closely as seen above. Nonetheless, the model actually fits the observed Q2, Q3, and Q4 data points fairly well (note the vertical distance between the actual observations and the model predictions), and the related parameter estimates survived a number of robustness tests quite well.
So, I think it’s appropriate to regard the model as reasonably accurate and reliable. Although as is always the case, there are perhaps other models that might be adequate or even better. But I think this model is sufficient for our purposes: Explaining and predicting Brazilian private consumption spending growth.
Relative marginal effect magnitudes of causal factors. I believe the marginal effect interpretations presented above are probably adequate to understand the main econometric results, but to continue with a theme I discussed in Part 2 of the series I think it will be helpful to summarize relative marginal effect magnitudes by their (approximate) relationship to Brazil’s internal or external economic sectors:
The above expression shows the aggregated marginal effects of the causal factors, where the marginal effects of NMDSR and DSR are shown as a difference between the two for simplicity. The effects are then aggregated into internal versus external sector effects (see Part 1) and then each is divided by the total sum of marginal effects to derive the relative effect from each sector, which sums to 100%.
With the caveat that it’s not exactly accurate to attribute a particular variable to Brazil’s internal or external economic sector–there are interrelationships–I think the above categorization of the relative marginal effects is helpful. Noting that seasonal effects are omitted, the above expression suggests that during the 2005-2016 sample period …
(i) Brazil’s consumption growth was mainly attributable to its internal economic sector factors, and
(ii) Brazil’s consumption contraction was mainly attributable to its external economic sector factors.
If true, that is both interesting and provocative. In any case, these are complex issues that need to be studied in a more integrated way than I’m doing here, so I will leave any such issues for future articles and end things here.
3. Preliminary conclusions: Brazilian private consumption
What have we learned about Brazil’s economic downturn by studying its private sector consumption spending? I think the econometric model and related analyses of 2005-2016 data are sufficient to allow us to preliminarily conclude …
(1) Employment growth, lending, household debt levels, exchange rate changes, and export growth all had significant effects on private consumption growth.
(2) Similar to findings of my brief study of Brazilian employment growth, a mid-2012 contraction in household non-mortgage debt seems to be–or is associated with–leading economic indicator(s) of Brazil’s broader economic downturn in mid-2014.
(3) Most interestingly, similar to results from the study of employment growth, external sector economic factors seemingly had a more important effect than would reasonably be expected given Brazil’s economic history; and, again, the effect appears to be negative.
I will continue to examine these issues in my next article … vai em frente!
Caveats. Please note: (i) views presented above are my own and do not reflect those of others; (ii) like anyone, I’m not infallible and am responsible for any errors; (iii) I greatly appreciate being informed of any significant errors in facts, logic, or inferences and am happy to give credit to anyone doing so; (iv) the above article is subject to revision and correction; and, (v) the article cannot be construed as investment or financial advice and is intended merely for educational purposes. MMc