Friedman's Permanent Income Hypothesis

The puzzle Friedman set out to solve

By the 1950s, two stylized facts about consumption refused to fit one Keynesian equation. In a single year's cross-section of households, the rich saved a larger fraction of income than the poor — the marginal propensity to consume (MPC) fell with income. But in the long-run aggregate time series, the U.S. saving rate had been almost flat for a century even as per-capita income grew tenfold. Both observations cannot come from a stable consumption function C = a + b·Y.

Simon Kuznets noticed this contradiction in 1942; it became known as the Kuznets puzzle. Friedman's 1957 book A Theory of the Consumption Function resolved it. The trick: distinguish the long-run resources households actually plan around — permanent income — from the volatile paycheck arriving this period.

The model

Friedman split measured income Y into two components, permanent (Y^P) and transitory (Y^T):

Y = YP + YT

Permanent income is the steady, expected return on a household's lifetime resources — labor earnings, human capital, financial wealth. Transitory income is the deviation: a bonus, an inheritance, a lost work week, a hailstorm-damaged car.

Consumption is similarly split: C = C^P + C^T. The hypothesis claims three things. First, planned permanent consumption is proportional to permanent income, with a coefficient that depends on preferences, interest rates, and time horizon but not on the level of Y^P:

CP = k · YP

Second, transitory consumption and transitory income are essentially uncorrelated — Friedman's strong assumption. Third, when we run a regression of measured C on measured Y, the apparent MPC is a weighted average that depends on how much of the variation in Y is permanent vs transitory:

C = β · YP + (1 − β) · YT
   ≈ 0.95 · YP  +  0.05 · YT

In a cross-section, much of the income variation across households is permanent (some really are doctors, others cashiers), so the estimated MPC is high. In a single year's time series, most variation is transitory (a recession, a tax surprise), so the estimated MPC is low. Same households, same preferences — different apparent MPC. Puzzle resolved.

Worked example — a $1,200 tax rebate

Imagine a household earning $60,000 per year in permanent income. The government sends a one-time $1,200 rebate cheque. What happens to consumption?

Under a naïve Keynesian function with MPC = 0.8, consumption rises by $1,200 × 0.8 = $960 this year. Under the PIH, the cheque is purely transitory. Plugging it into the formula:

ΔC = 0.95 · 0  +  0.05 · 1200  =  $60

Only $60 gets spent immediately. The other $1,140 is saved — building precautionary balances, paying off credit-card debt, or shoring up the retirement account. If the household lives for another 40 years and discounts at 3%, the $1,200 spread over the lifetime amounts to roughly $52 per year in permanent consumption gain. The PIH says households recognize this and adjust accordingly.

Now consider a permanent raise from $60,000 to $63,000. Under the same coefficients:

ΔC = 0.95 · 3000  +  0.05 · 0  =  $2,850

Almost all of the raise gets consumed, year after year. The MPC out of permanent income is twenty times the MPC out of transitory income. This asymmetry is the heart of the model.

Hall's random-walk extension

Friedman approximated expected permanent income with an exponentially weighted moving average of past income. Robert Hall (1978) replaced that ad-hoc rule with rational expectations and got a much sharper prediction. Under rational expectations and quadratic utility:

Ct+1  =  Ct  +  εt+1

Consumption is a martingale. Only news — innovations in expected lifetime resources — moves it. Anticipated income changes don't, because households already smoothed against them. The empirical bite: if you see somebody's consumption jump when a predictable bonus arrives, that's evidence against rational-expectations PIH. Hall tested it on U.S. data, finding lagged income had some explanatory power for consumption beyond lagged consumption — a small but significant violation of the random-walk benchmark. Subsequent work has refined the test rather than abandoning the framework.

PIH vs other consumption theories

	Keynesian (1936)	Permanent Income (Friedman 1957)	Life-Cycle (Modigliani 1954)	Random-Walk (Hall 1978)	Buffer-Stock (Carroll 1997)	Two-Agent (Mankiw 2000)
Determinant of C	Current income Y	Permanent income YP	Lifetime resources	Lifetime resources + news	Lifetime res. + precaution	Half permanent, half hand-to-mouth
MPC out of transitory	0.7–0.8	≈ 0.05	1/T (remaining life)	≈ 0	≈ 0.2–0.4	≈ 0.5
Consumption volatility	≈ Income volatility	Much smoother	Smoother across life	Random walk	Smoother, kinked	Mixed
Solves Kuznets puzzle	No	Yes	Yes	Yes	Yes	Yes
Predicts rebate stimulus	Strong	Weak	Weak	Zero	Moderate	Moderate
Empirical fit (rebates)	Too strong (~0.6)	Too weak (~0.05)	Too weak	Far too weak	Closer	Closest

The evidence — close, but not exact

Decades of rebate-cheque studies have stress-tested the PIH. Three influential rounds:

• Wilcox (1989): Predictable Social Security benefit increases moved consumption on the day of the cheque, not the day of the announcement. Violates Hall random-walk; consistent with liquidity constraints.
• Parker, Souleles, Johnson, McClelland (2006, 2013): Used randomized rebate-timing in 2001 and 2008 to identify the causal MPC out of transitory income. Found three-month MPC of roughly 0.25–0.40 on non-durables, and 0.50–0.90 once durables are included. Far above PIH's 0.05.
• Kaplan and Violante (2014): Showed that the high empirical MPC is concentrated among "hand-to-mouth" households who hold liquid wealth below one paycheque. Households with even modest liquid savings behave PIH-style. About 1/3 of U.S. households were hand-to-mouth at the time of the study; many were wealthy in illiquid assets (housing, retirement accounts).

The synthesis: PIH describes how an unconstrained household with full access to credit would behave. About two-thirds of U.S. households roughly do. The rest are constrained and behave Keynesian, which is why rebate cheques still generate measurable demand.

Counterarguments

The excess-sensitivity puzzle. Predictable income changes shouldn't move consumption under rational-expectations PIH, but they do. Excess sensitivity has been documented for tax refunds, paychecks, Social Security checks, and even the geographic shifting of payday across months. Magnitudes are small but statistically robust.

The excess-smoothness puzzle. Conversely, when income shocks are permanent, consumption sometimes moves less than the PIH predicts (Campbell and Deaton 1989). Households appear cautious about updating their long-run forecasts, consistent with learning or buffer-stock motives.

Habit formation. If utility depends on consumption relative to a habit stock, even unconstrained agents smooth less aggressively than pure PIH. Habit models fit asset-pricing data (Campbell-Cochrane 1999) and consumption growth volatility better than pure CRRA preferences.

Precautionary saving. When future income is uncertain and utility has positive third derivative (prudence), households save more than the certainty-equivalent PIH predicts. The buffer-stock model embeds this: there's a target wealth-to-income ratio households defend against bad draws, generating a kinked consumption rule that looks PIH-like at high wealth and Keynesian at low wealth.

Common pitfalls

• Confusing the level effect with the marginal effect. A high consumption level doesn't imply a high MPC. PIH households with high lifetime resources have high C but still respond little to transitory shocks.
• Treating Y^P as observed. Permanent income is an unobservable expectation. Empirical work either approximates it with smoothed income, instruments out the transitory part, or estimates the full income process.
• Assuming all households are unconstrained. A third of U.S. households cannot smoothly borrow against future income; PIH doesn't describe them. The two-agent model is now the more common workhorse for fiscal-policy analysis.
• Reading "rational" as "knows the future". Rational expectations PIH says agents form unbiased forecasts. They still make errors; those errors become the random-walk innovation ε.
• Forgetting the discount-rate term. The MPC on permanent income k depends on the real interest rate and the planning horizon, not just on preferences. Calibrations differ by these assumptions.