### Appendix

**Statistical framework mapping individual-level financial circumstances to investing**

To better understand the links between income growth—as emphasized in this piece—and potentially other financial factors, like a person’s liquid cash buffer, we use the regression framework described below. Predictors of investment transfers tested here include a person’s average income level, recent income growth, cash buffers (defined as cash liquidity divided by average monthly spending, relative to an individual-level baseline), and other controls. Figure A.1 shows the regression results broken out by three specifications across two samples: the full sample and a separate, “investor only” sample, which includes data of individuals with at least one prior investment transfer. The same methodology for income growth—both “short-term” and “sustained”—used in Finding 2 apply here and are additionally applied to expense growth, to incorporate a potential role for savings.

The simplest specification—Model 1—includes income growth (both short-term and sustained) cash buffer (deviation in t-1 from the average over the t-24 to t-13 baseline) in addition to month fixed effects to capture seasonality. In the full sample, the estimates imply a one standard deviation increase in income growth, in a single month, increases the likelihood of a transfer to investments in that month by about 15 percent. Given that investment transfers are infrequent (i.e., 3 percent of the more than 1 billion person-months in the data), this amounts to just a 0.5 percentage point increase in the predicted probability of investing in the month.^{21} Sustained income growth adds another 0.4 percentage point increase in the probability, bringing the increase in likelihood of investing from 1-standard deviation of both short-term and sustained income growth to 28 percent. The role of income growth in predicting investment behavior is notably stronger when expense growth is low relative to income growth—that is, when an individual’s savings rate is increasing—which is shown in Models 2 and 3, which include measures of growth in account outflows excluding transfers, a proxy for spending. Model 3 includes the variables in Model 2 plus age and time controls. These do not notably affect the coefficients of interest.

The empirical relevance of income growth, saving, and cash buffers is several times stronger in terms of percentage point contributions to investment likelihood when limiting the sample to individuals that have at least one prior investing transfer—i.e., after removing the individuals that are unlikely to become investors, for any reason. For example, in Model 1, the combined effects of a one standard deviation increase in one month’s income, on top of a one-standard deviation increase in sustained growth is almost 4 percentage points (almost 10 times as large as the full sample effect). Additionally, cash buffer increases—which have limited explanatory power in the full sample—have a notable role among prior investors, who may use external accounts as an outlet for excess cash.