In the past few years, a persistently low level of consumer sentiment, as measured by the University of Michigan, Conference Board, or Gallup surveys, relative to conventional economic measures has puzzled analysts, including this one [1]<\/a> [2]<\/a> [3]<\/a> [4]<\/a> [5]<\/a>. This puzzle is illustrated by the evolution of the U. Michigan series (FRED series UMCSENT), and the fitted values using 2011M01-2024M08 data on unemployment and year-on-year inflation.<\/p>\n \u00a0<\/p>\n Figure 1:<\/strong> University of Michigan Consumer Sentiment index (bold black), fitted values using unemployment and inflation (see text) (tan), +\/- 1 standard error band (gray line). NBER defined peak-to-trough recession dates shaded gray. Source: University of Michigan via FRED, NBER, and author\u2019s calculations.<\/em><\/p>\n Here\u2019s the regression results (where inflation is 12 month growth rate, in %):<\/p>\n UMCSENT<\/em> = 104.12<\/strong> \u20132.455<\/strong> UNRATE<\/em> \u2013 3.727<\/strong> INFL<\/em><\/p>\n Adj-R2 = 0.51, SER 9.256, DW = 0.23, N=164, Sample 2011M01-2024M08. Bold denotes significance at 10% level using HAC Newey-West standard errors.<\/p>\n Peak gap is in mid-2023 and end-2023.<\/p>\n Partial resolution is shown in the following picture, motivated by Bolhuis et al (2024)<\/a>.<\/p>\n Figure 2:<\/strong> University of Michigan Consumer Sentiment index (bold black), fitted values using unemployment and inflation\u00a0 (tan), using unemployment, inflation, and 30 year mortgage rate (green). NBER defined peak-to-trough recession dates shaded gray. Source: University of Michigan via FRED, NBER, and author\u2019s calculations.<\/em><\/p>\n The green line is the fit using the baseline equation augmented with 30 year mortgage rates.<\/p>\n UMCSENT<\/em> = 135.67<\/strong> \u20133.669<\/strong> UNRATE<\/em> \u2013 3.328<\/strong> INFL \u2013 <\/em>5.97<\/strong> MORTGAGE30US<\/em><\/p>\n Adj-R2 = 0.73, SER 6.785, DW = 0.48, N=164, Sample 2011M01-2024M08. Bold denotes significance at 10% level using HAC Newey-West standard errors.<\/p>\n Where MORTGAGE30US is the FRED series for the 30 year mortgage rate in the US. Note that in both regressions, I\u2019ve used HICP (which does not include housing costs) as my price index.\u00a0 The mortgage rate is almost as important as the unemployment rate, according to standardized (beta) coefficients (0.53 vs. 0.58, and 0.62 for inflation).<\/p>\n These results indicate that a 1.1 percentage point decline in MORTGAGE30US since the October 2023 peak (thru August) has accounted for an approximately 6.5 point increase in UMCSENT.<\/p>\n Similar results are exhibited by the Conference Board\u2019s Consumer Confidence index (although HICP y\/y inflation doesn\u2019t show up as statistically significant):<\/p>\n Figure 3:<\/strong> Conference Board Consumer Confidence Index index (bold black), fitted values using unemployment and inflation (tan), using unemployment, inflation, and 30 year mortgage rate (green). NBER defined peak-to-trough recession dates shaded gray. Source: Conference Board via Investing.com, NBER, and author\u2019s calculations.<\/em><\/p>\n The regression results are:<\/p>\n CONFIDENCE_CONFBD<\/em>\u00a0= 175.28<\/strong> \u20139.544<\/strong> UNRATE<\/em> \u2013 0.716\u00a0<\/strong>INFL \u2013 4.984<\/strong><\/em>\u00a0<\/strong>MORTGAGE30US<\/em><\/p>\n Adj-R2 = 0.63, SER 13.640, DW = 0.40, N=164, Sample 2011M01-2024M08. Bold denotes significance at 10% level using HAC Newey-West standard errors.<\/p>\n Why is this only a possible resolution? If one extends the sample before 2011M01, the regression results fall apart. Inclusion of MORTGAGE30US then yields a positive coefficient, insignificant at first, and significant as the sample extends further back in time (see e.g. 1991M01-2024M08; one can use CPI ex-shelter as a longer series for prices). Since this instability in the relationship is not associated with the change in including housing costs in the CPI as noted by Bolhuis et al (2024)<\/a>, I think of this as a remaining puzzle. Perhaps if housing prices were also included, this would remedy the problem (actually, I have checked \u2013 it kind of does).<\/p>\n Finally, note that mortgage rates enter asymmetrically into UMCSENT for Democrats vs. Republicans. Mortgage rates have an insignificant impact for Democrats\u2019 sentiment. A Trump dummy dominates for both Republicans and Democrats, in terms of influence as measured by a standardized coefficient. Hence, the inclusion of mortgage rates do not explain all the seemingly anomalous behavior of sentiment indices.<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n \u00a0<\/p>\n<\/p><\/div>\n![\"\"](\"https:\/\/econbrowser.com\/wp-content\/uploads\/2024\/09\/umcsent_wo1.png\")
<\/a><\/p>\n![\"\"](\"https:\/\/econbrowser.com\/wp-content\/uploads\/2024\/09\/umcsentfit_wwo_int-1.png\")
<\/a><\/p>\n![\"\"](\"https:\/\/econbrowser.com\/wp-content\/uploads\/2024\/09\/confidence_wwo_int.png\")
<\/a><\/p>\n