Do Consumers React to Anticipated Income Changes? Evidence from the Alaska Permanent Fund

Questions & answers

Q1. What does the certainty-equivalent LC/PIH predict, and what prior puzzle motivates this paper?

The certainty-equivalent version of the life-cycle/permanent-income hypothesis — the version in which the expected variance of consumption is constant — implies that consumers should not respond to predictable changes in their income (Introduction, fn. 1). Hsieh notes that recent natural-experiment tests of this prediction had instead found the opposite: Parker (1999), using the cap on earnings subject to the Social Security tax, and Souleles (1999), using income tax refunds, both found that consumption is “excessively sensitive” to anticipated income changes. Hsieh frames two competing explanations for that pattern — either the canonical consumption model is wrong, or the anticipated changes those papers study are simply too small and irregular for households to bother incorporating into their plans — and sets up the Alaska Permanent Fund as “an unusually powerful test” because its payments are large and easy to anticipate, which should let the two explanations be told apart.

Q2. Why are Alaska Permanent Fund Dividend payments an especially clean test case?

The PFD payments are large, arguably exogenous to the households that receive them, and clearly anticipated, since every resident must file an application by the end of March — about six months before the October payment — to qualify (Sec. I). A family of four received $6,164 from the Fund in 1998. Although the final official announcement of the payment size comes in late September, the formula used to calculate it from the Fund’s investment earnings is widely known and publicized monthly, so Alaska newspapers routinely publish accurate estimates of the payment before the official announcement (the paper quotes two Anchorage Daily News headlines forecasting the dividend, one of which anticipated the actual $1,131 figure for 1996 under the headline “‘96 Dividend Could Reach $1,100 Mark”). Because the amount and October timing of the payment are both precisely known to the researcher (every household received its payment in the fourth quarter after 1984), Hsieh can construct an exact measure of the anticipated income change rather than relying on proxies.

Q3. What is the main empirical specification, and what does it find for nondurable consumption?

Hsieh estimates a linear Euler-equation specification in which the log change in household consumption from the third to the fourth quarter is regressed on the PFD payment per person times family size, scaled by family income, with the LC/PIH predicting a coefficient of zero (Sec. III, eq. 1). Using CEX data on roughly 800 Alaskan households per year from 1980–81 and 1984–2001, the point estimate of the consumption-income elasticity is positive but economically and statistically insignificant across three specifications (pooled, with year dummies, with family-size controls): a 10-percent increase in household income from the dividend is associated with only a 0.002 percent increase in nondurable consumption (Table 2, Sec. III) — even though the dividend raised the typical household’s quarterly income by just over 20 percent.

Q4. What about durable consumption, and what pattern shows up there?

The coefficient estimates for durable-goods expenditure are small but marginally significant, and — surprisingly — negative: the change in durables spending from Q3 to Q4 is smaller the larger the dividend payment, which Hsieh reads as evidence that households buy durables in the third quarter in anticipation of the October payment (Table 2, Sec. III), rather than waiting until after receiving it. This is consistent with, not contrary to, forward-looking anticipation of the payment.

Q5. What corroborating evidence does the paper offer on saving and debt behavior?

Using the CEX’s first- and fourth-interview questions on debt and account balances (exploiting the random timing of a household’s own survey start date across the year), Hsieh estimates that a representative Alaskan household’s net assets rose by about $1,760 in October — comprising roughly $680 less consumer debt and $1,080 more in savings/checking balances — which is close to the roughly $2,000 average Fund payment that quarter (Sec. III, Fig. 1). This supports the Table 2 regression finding that households are saving or paying down debt with the dividend rather than spending it, and he supplements it with kernel-density evidence (Fig. 2) that the distribution of the Q3-to-Q4 consumption change in Alaska does not look different from that of the other 49 states.

Q6. Does the paper rule out alternative explanations such as fewer liquidity constraints or a different consumption composition in Alaska?

Hsieh considers and rejects three alternative explanations for the smoothing result. First, a composition effect — Alaskans spending disproportionately on food, whose consumption is harder to shift — is ruled out because Alaskan households spend about the same food share of nondurable expenditure (37 percent) as the rest of the US (39 percent), and splitting nondurable consumption into food, entertainment, and apparel subcategories (Table 4) shows no large differential response in any category. Second, weaker liquidity constraints or less buffer-stock saving among Alaskans is not supported: the average Alaskan household head is seven years younger than elsewhere (which, given that younger households track income more closely, would predict more buffer-stock behavior, not less), and restricting the sample to below-mean-income households (Table 5) still yields small, insignificant responses. Third, Hsieh considers that the PFD is simply a better-measured anticipated-income instrument than those used elsewhere, rather than evidence that households generally succeed at consumption smoothing.

Q7. What is the paper’s key within-household comparison, and why is it the central piece of evidence?

Restricting the sample to households with income-tax-refund data, Hsieh finds that the same Alaskan households who smooth their PFD payments are, by contrast, “excessively sensitive” to their income tax refunds: a refund that raises income 10 percent in the second quarter raises nondurable consumption by about 3 percent, a statistically significant effect, while the PFD response for this same restricted sample remains small and statistically indistinguishable from zero (Sec. IV, Table 6). Because this is a within-household, same-survey-period contrast, Hsieh argues it isolates the type of anticipated income change — not some fixed household trait — as the source of the earlier literature’s “excess sensitivity” findings, and reads it as evidence for bounded rationality: households incorporate anticipated income changes into consumption only when the changes are “large, regular, and easy to predict,” but not when they are “small and irregular” (Sec. V).

Q8. What does the paper conclude about the implications for fiscal policy?

Hsieh concludes that the aggregate consumption response to anticipated fiscal-policy income changes depends on whether the policy change is transparent and has a significant, easily understood effect on a typical household’s income (Sec. V) — not simply on whether the change is technically foreseeable in advance. He is explicit that this finding of LC/PIH-consistent behavior for the Permanent Fund does not imply that fiscal policy generally fails to affect aggregate consumption: “many tax and fiscal policy measures will probably have an effect on aggregate consumption as long as people find it difficult and costly to understand precisely how their incomes are affected by these policies.”