Corrigendum for Dividend Dynamics, Learning, and Expected Stock Index Returns

• DOI: http://doi.org/10.1111/jofi.12786
• Date: 01 August 2019
• Published date: 01 August 2019

THE JOURNAL OF FINANCE •VOL. LXXIV, NO. 4 •AUGUST 2019

Corrigendum for Dividend Dynamics, Learning,

and Expected Stock Index Returns

RAVI JAGANNATHAN, BINYING LIU, and JIAQI ZHANG∗

We discovered inconsistencies in our coding for Jagannathan and Liu (2019)

that, after addressed, has led to changes in the tables and figures that we

reported. These changes do not in anyway affect any of the paper’s statements,

findings, or conclusions. We report updated tables and figures in this erratum

and highlight any statistics where the change is nontrivial by underlining it.

The inconsistencies are as follows:

rIn the paper, we mention that our Kalman filter estimation is based on

nonoverlapping annual dividend data. However, in the codes we used a

mixture of nonoverlapping and overlapping regressions to estimate param-

eters and state variables in the AR[1] processes of earnings-to-dividend

ratios and inflation rates. Here, we correct this inconsistency, and use

nonoverlapping data throughout all parts of the paper.

rIn Tables III and Vin Section I of the paper, we had accidentally lagged the

right side variable by six months while estimating the model. This error

has now been corrected.

rIn Figure 7of the paper, we forgot to specify, when reporting summary

plots for earnings-to-dividend ratios and inflation rates, that these plots

were for the variables de-meaned. Here, we report the figure without de-

meaning.

rIn estimating expected returns from the long-run risks model, we acci-

dentally mistook inflation lagged by one year as current inflation in some

parts of our coding. This has been corrected.

The nontrivial changes are as follows. In Table III (and Table V), the out-of-

sample R2for our dividend model drops from 0.413 (0.395) to 0.320 (0.331), but

remains statistically higher than the corresponding R2s of competing models.

Return predictability R2for the full learning model in Table IX increases from

0.271 to 0.291 for the full data sample, in Table XI it increases during expan-

sions from 0.191 to 0.252, and decreases during recessions from 0.641 to 0.474.

These changes do not change the main conclusions in the paper. The Internet

∗Ravi Jagannathan is affiliated with Kellogg School of Management, Northwestern University,

and NBER, ISB, SAIF. Binying Liu is with the Hong Kong University of Science and Technology.

Jiaqi Zhang is affiliated with Kellogg School of Management, Northwestern University.

DOI: 10.1111/jofi.12786

2107