One of the enduring mysteries of the past two years has been why measured economic sentiment has remained so subdued despite objectively good economic outcomes.
figure 1: Summation of misery indices for unemployment and inflation, % (black, left scale), and Michigan consumer sentiment, reciprocal (tan, right scale). Inflation rate is the year-on-year growth rate of CPI. Recession peak-to-trough dates as defined by NBER are grayed out. Sources: BLS, U.Michigan, via FRED, NBER, author’s calculations.
One thing to realize is that there are actually three distinct groups in the Michigan sample. Categorical results—Democrat/Democrat Lean, Independent, Rep./Rep. Lean—are reported only occasionally, though more frequently in recent years.
figure 2: Michigan Democrat/Democrat Consumer Sentiment (blue), Rep./Republican Consumer Sentiment (red). Light orange shading indicates the Trump administration. Recession peak-to-trough dates as defined by NBER are grayed out. Sources: University of Michigan, NBER, author’s calculations.
Note that with the end of the Trump administration and the arrival of Biden, there has been a marked shift in sentiment that does not appear to be related to changes in general economic conditions (although this is not to say that inflation and unemployment do not affect measurable sentiment ).
This is confirmed by the following regressions (2008M01-2023M06) for Democrat Sentiment and Rep. Sentiment:
sendthe democratic party = 103.33 – 1.92you – 1.85PI -28.76GW Bush – 13.02trump card + 1.52biden – 38.6Coronavirus
adjust-R2 = 0.48, SER = 9.046, DW = 0.42, Nobbs = 109. bold (emphasize) indicates significance at 10% msl (15%) using the HAC robust (Newey-West) standard error.
Where send is the University of Michigan Sentiment Index, you is the unemployment rate, PI is the 12-month CPI inflation rate, Covid is the death rate per 1,000 people for 2020M01-2021M06, and 0 otherwise. GWBush, Trump, and Biden are dummy variables for each administration (Obama is the dummy variable that was omitted, so the constant can be interpreted as an average holding everything else constant during the Obama administration.)
sendrepresent = 102.71 – 4.03you – 1.62PI -3.84GW Bush + 34.64trump card – 26.87biden + 27.4Coronavirus
adjust-R2 = 0.91, SER = 9.16, DW = 0.70, Nobbs = 109. bold Indicates significance at 10% msl using HAC robust (Newey-West) standard errors.
Interestingly, from a statistical significance point of view, the sentiment of these two groups has not been affected by the transaction per capita of the new crown virus.
By the numbers, Democrats do see the Bush and Trump administrations as negative for sentiment, while Republicans have a huge plus for Trump and a huge negative for Biden.
If Democrat/Democratic leaning and Rep. Then one would not expect a differential effect. However, the Rep/Rep slope Trump coefficient is almost 3 times (in absolute value) the Dem/Dem slope coefficient. Furthermore, while Biden’s Dem/Democrat skew is positive, its magnitude (in absolute value) (1.5) is dwarfed by the Rep/Republican skew of 26.9 (!).
Thus, while Rep/Rep thinning has a weight of about 24% and Dem/Dem thinning has a weight of about 34% (about 46% Ind), the intensity of the Rep/Rep thinning sensation explains a large portion of the overall measurement.
Another way to see who chairs has a huge impact on Rep. Lean sentiment is to look at the 12-month change in UMCSENT and break it down into the percentages for each group.
image 3: Michigan August-August consumer confidence change (black squares), with contributions from Democrat/Democrat leaning (blue bars), Rep/House leaning (red bars), and Independents (grey bars). Light orange shading indicates the Trump administration. Source: University of Michigan, author’s calculations.
Figure 3 highlights a major shift in House sentiment associated with the Trump inauguration in 2017, and another major shift (in the opposite direction) associated with the Biden inauguration in 2021. While Democrat/Democrat leanings contributed a lot, they were not as large as Republican shifts. The huge effect on reps is even more impressive when one recalls that reps/reps lean body mass was 24% (vs. dem/dem lean body mass was 34%).
