Update: I found a survey that addresses exactly the same questions as outlined in this post (even categorizing shocks and responses similarly!).

I am interested in how individuals respond to changes in their income. The changes may affect periodic income, temporarily or permanently, in the form of income shocks, such as layoffs, or be less egregious periodic income modulations such as delayed pay-days, bonuses and raises. They may also be completely independent from periodic income, but expected, such as receiving your tax refund. Finally, they may be completely unexpected financial windfalls, such as having your security deposit returned by the slumlord you used to rent from in darker times.

Clearly, there is heterogeneity in what income is modulated, and how this income is modulated. I categorize effects on income as follows:

1. Duration of effect: Permanent vs transient.
2. Direction of effect: Positive vs. negative.
3. Magnitude of effect: Large vs. small.
4. Target of effect: Primary income vs. none.

There is also heterogeneity in how different individuals respond to such changes in income; some do not respond at all, while some increase spending to unsustainable levels. I categorize response to income (assumed to be observed as changes in spending) as follows:

1. Duration of response: Permanent vs. transient.
2. Direction of response: Positive vs. negative.
3. Magnitude of response: Large vs. small.
4. Target of response: Overall spending vs. specific product categories.

These categorizations are clearly coarse-grained: responses may be highly non-linear, for example, starting out highly positive, then correcting themselves and going negative, and finally stabilizing at zero. It is nevertheless useful to have some simple way to categorize the wealth of research findings on this topic.

Note that I restrict this study to that of income; hence, I assume away any effects of non-income losses such as being the unfortunate victim of a street robbery. Such events would definitely show up as expense fluctuations, so I may be better served by incorporating hidden causes of expense modulation into my model.

Why should we care about how people respond to income? The first motivation is to identify individuals who may harm themselves financially in response to income shocks (I read that over 70% of lottery winners go bankrupt). Once identified, planned income shocks to such individuals (such as the disbursement of tax refunds or bonuses) could be spread out over time to alleviate adverse responses. The second motivation is to guide setting the frequency of pay in situations where there is a choice; my hypothesis is that different pay frequencies lead to different levels of welfare for each individual, given that they respond to pay-days in different ways. The third motivation is to better predict any given individual’s future expense transactions (time and amount) from past transactions.

What I have in my data are the times and amounts of income, the times and amounts of expenses and a coarse product category for each purchase. I am also reasonably certain that this data forms a “financial sensor log”: an involuntary and automatic stream of all the income and expense transactions of every individual under study. This sets it apart from the previously studied streams of voluntary actions, such as Foursquare check-ins and tweets, where multiple layers of choice (the choice to eat, the choice to use Foursquare, the choice to check-in) add barriers between observing the activity and making claims about natural human behaviour; as a simple example, are Facebook “friends” friends? I think this difficulty in making claims on human behaviour from such samples of observed activity is called selection bias.

In the next post, I will survey a number of papers studying the income elasticity of consumption. The seeds of this literature review were planted by Alan Montgomery over at Tepper and Brian Kovak who taught me something awesome from 9.00am - 10.30am every morning this semester.