Consumption Demand

The empirical study of aggregate consumption behavior has long been shaped by a central puzzle. Early Keynesian theory posited an absolute income hypothesis, in which real aggregate consumption expenditures are a stable function of current real disposable income: C = a + bY, where b is the marginal propensity to consume. This specification implies that the average propensity to consume, C/Y = a/Y + b, declines as income rises. Cross-section surveys of household consumption and income indeed revealed such a falling APC, with the MPC well below unity and lower than the APC at high income levels. However, long-run time-series data over many decades, notably Kuznets's finding that the aggregate saving rate was remarkably stable in the United States, showed a constant or even slightly rising APC, implying a proportional long-run consumption function.

Reconciling this discrepancy required moving beyond the mechanical link between current income and consumption. The two most influential microeconomically-founded theories that succeeded in explaining the short-run/long-run divergence are the permanent income hypothesis of Milton Friedman and the life-cycle hypothesis of Franco Modigliani, Richard Brumberg, and Albert Ando. Their common core is the idea that rational, forward-looking households make consumption decisions on the basis of their lifetime resources, not merely their current income. The modern restatement of this idea through the intertemporal optimizing framework (the Euler equation approach) further sharpens the distinction between short-run and long-run consumption responses.

In cross-sectional budget studies and in short-run time-series regressions, the relationship between consumption and current disposable income is well approximated by the Keynesian absolute income hypothesis with a positive intercept. Under this specification the APC declines monotonically with income: richer households and richer periods save a larger fraction of income than poorer ones. This was precisely the pattern Keynes anticipated. Yet over horizons of decades the APC was approximately constant despite a massive increase in real per-capita income. In other words, the long-run consumption function appeared to be a ray through the origin, C = kY, while the short-run function exhibited a positive intercept and a substantially smaller slope. Any satisfactory theory of consumption demand must explain why the same economy generates two such different functional relationships depending on the time horizon over which data are aggregated.

The modern theory reinforces the central message: the short-run consumption function—what one would estimate in regressions of current consumption on current income—is inherently unstable, because its parameters depend on the stochastic process of income and on the information used by households. The long-run relationship, however, approaches a proportionality between consumption and permanent (or lifetime) income, consistent with the empirically stable saving rate observed over decades.

The relative income hypothesis, advanced by James Duesenberry, marked a decisive break from the Keynesian absolute income hypothesis by grounding consumption behavior in social interdependence and irreversible habits. Duesenberry argued that consumption is a function of a household's income relative to the income of others and to its own past peak income. This dual relativity generates two key mechanisms: the demonstration effect and the ratchet effect.

Duesenberry's starting point is that individual utility depends not on the absolute level of consumption but on consumption relative to the social norm. The household's optimization yields a consumption function where consumption depends on both own income and average income in the community. At the individual level, the marginal propensity to consume out of own income is lower than the MPC out of social income, since part of consumption is driven by keeping up with the Joneses. Aggregating over all households, the economy-wide ratio of consumption to income becomes a function of the distribution of income: a more equal distribution reduces the weight of high-saving households and raises the aggregate consumption-to-income ratio.

The second pillar is the ratchet effect: a household's consumption is not only relative to others', but also relative to its own past highest level of consumption. If income falls, consumption is slow to adjust downward because households strive to defend the living standard previously achieved. Duesenberry's aggregate specification is C/Y = a + b(Y/Yp), where Yp is previous peak income. This formulation captures the observation that during a recession, consumption falls proportionally less than income, while in a boom it rises proportionally less, making the short-run MPC lower than the long-run APC.

The relative income hypothesis was explicitly designed to reconcile three observed regularities that plagued the absolute income theory. Cross-section budget studies consistently show that high-income households save a larger fraction of their income than low-income households. Short-run time series reveal a relatively stable APC over the cycle—consumption is less volatile than income, and the short-run MPC is lower than the long-run MPC. Long-run data show a remarkably constant APC of about 0.9 despite massive growth in real income. Duesenberry's resolution: in the short run, the ratchet effect holds consumption up during downturns and restrains the increase during upswings, generating a low short-run MPC and a near-constant APC. In the long run, as income grows over decades, the previous peak income is continuously revised upward, so the constraint binds less and the aggregate APC remains roughly constant. Simultaneously, as all households' incomes rise proportionately, the consumption norm also rises, making the APC independent of the absolute level.

Comparisons with the permanent income hypothesis and the life-cycle hypothesis are instructive. All three theories predict a roughly constant long-run APC and a lower short-run marginal propensity to consume. However, the relative income hypothesis does not rely on rational forward-looking planning or on a strict distinction between transitory and permanent income. Instead, it derives the short-run-long-run reconciliation from habit persistence and social interaction. Modern habit-formation models in which utility depends on current consumption relative to lagged consumption can be seen as formal descendants of the ratchet idea, albeit with micro-optimizing agents. The relative income hypothesis also foreshadows the recent macro literature on consumption externalities and "keeping up with the Joneses," where marginal utility depends on own consumption relative to aggregate consumption.

From a behavioral perspective, the relative income hypothesis implies that consumers' welfare gains from income growth are partially offset if others' incomes grow as well, because the reference point shifts. This helps explain the Easterlin paradox: average self-reported happiness does not rise with per capita income over time. In macroeconomic policy, the ratchet effect suggests that temporary tax cuts may have muted effects on consumption if households perceive them as transitory relative to their past peak income. Conversely, a severe recession that permanently reduces the peak income norm may lead to a protracted period of low consumption even if incomes recover.

Milton Friedman's permanent income hypothesis, presented in his 1957 A Theory of the Consumption Function, provided the first rigorous micro-founded reconciliation of the Kuznets puzzle. Friedman's central insight was that observed income and observed consumption in any given period are imperfect measures of the underlying economic variables that actually govern household decisions.

Friedman decomposed measured income and measured consumption into permanent and transitory components. Permanent income is defined as the expected, annuity-like flow of resources the household can count on over its planning horizon—essentially the return on total wealth, human and non-human. Transitory income captures unexpected windfalls, temporary layoffs, bonuses, and other one-off deviations. Analogously, permanent consumption is the planned, sustainable consumption flow, while transitory consumption reflects timing irregularities such as an unusually large medical bill or the replacement of a durable good.

The hypothesis rests on three orthogonality assumptions: permanent and transitory income are uncorrelated, permanent and transitory consumption are uncorrelated, and transitory income and transitory consumption are uncorrelated. The third condition is the most consequential: it asserts that transitory income shocks are neither consumed nor systematically saved in the same period—they are smoothed through saving and borrowing. The structural behavioral relationship is then posited between the permanent components alone: permanent consumption is proportional to permanent income, where the proportionality factor depends on the household's wealth-to-income ratio, tastes, and the horizon, but is independent of the level of permanent income. For a homogeneous population with common proportionality factor, this implies the long-run consumption function C = k Y^P, a ray through the origin with constant APC equal to k.

Suppose an econometrician, observing only measured income and consumption, runs a cross-sectional regression of consumption on income. Using the orthogonality conditions, the OLS slope converges to k times the fraction of income variance that is permanent. Because this fraction is less than one, the estimated short-run MPC is strictly smaller than the long-run MPC. In the long run, as transitory fluctuations average out, the fraction of variance that is permanent approaches one and the estimated MPC approaches k, recovering the Kuznets ray. The same algebra explains why the cross-sectional APC falls with measured income: high measured incomes contain a positive transitory component on average, and since transitory income is saved rather than consumed, the APC is depressed for high-income observations.

The PIH thus delivers a unified account: the apparent conflict between short-run and long-run evidence is a statistical artifact of measuring the wrong variables. Households do not mechanically consume a fixed fraction of current income; they consume a fixed fraction of permanent income and use saving as a buffer stock against transitory fluctuations. The PIH further predicts that windfall income—a large, purely transitory addition to current income—will be saved almost entirely, an implication tested and largely confirmed by studies of one-time dividend payments and tax rebates.

The life-cycle hypothesis, developed by Franco Modigliani and Richard Brumberg, arrives at closely related implications through an explicit intertemporal optimization framework. Where Friedman emphasized the statistical decomposition of income, Modigliani emphasized the demographic and horizon structure of earning and spending.

Consider a household that lives for T periods, derives utility from consumption in each period, and faces a sequence of labor incomes. With access to a perfect capital market at a real interest rate r and initial assets, the household maximizes lifetime utility subject to the lifetime budget constraint that the present value of consumption equals initial assets plus the present value of labor income—total lifetime wealth. The first-order conditions yield the celebrated Euler equation for consumption: the marginal utility of consumption today equals the product of the subjective discount factor, the gross interest rate, and the expected marginal utility of consumption tomorrow. When the interest rate equals the subjective discount rate and utility is quadratic, optimal consumption is perfectly flat: the household consumes the same amount in every period, regardless of the time profile of income.

Plugging constant consumption into the lifetime budget constraint gives optimal consumption as the annuity value of total lifetime resources. The annuity factor converts lifetime resources into a constant consumption stream. It is the marginal propensity to consume out of wealth. For long horizons, this factor is approximately the interest rate; for finite horizons, it exceeds the interest rate to exhaust all resources by the final period. The quantity is the household's permanent income—the maximum constant consumption that can be maintained without running down net worth below zero. This is the heart of Modigliani and Brumberg's contribution: consumption depends on total lifetime resources, not on this period's earnings.

The life-cycle hypothesis gets its name from the systematic pattern of labor income across the human life span. In a typical individual's life, youth income is low (or zero, while in education) but expected to rise. The household smooths consumption by borrowing; consumption exceeds income, so saving is negative. During prime earning years, income peaks. The household repays earlier debts and accumulates wealth for retirement; consumption is less than income, so saving is positive. During retirement, labor income falls to zero. The household dissaves, financing consumption out of previously accumulated assets; again consumption exceeds income, saving negative. The consumption path is flat, while income follows a hump-shaped profile. Consequently, wealth first declines, then rises, and finally falls, generating the characteristic hump-shaped wealth trajectory. The saving rate varies systematically with age: negative for the young, high for the middle-aged, negative for the retired.

An economy's aggregate consumption function depends on the demography and the distribution of wealth across cohorts. Growth in population or productivity that increases the share of young dissavers relative to older savers can alter the aggregate saving rate. Under the LCH, aggregate consumption can be expressed as a function of current labor income and total net worth. If we approximate human wealth as proportional to current labor income—which holds exactly if the economy is in a steady state with constant growth—we obtain a modified Keynesian-type consumption function where consumption depends on both wealth and current income. This formulation highlights that the MPC out of wealth is small because wealth is consumed over the entire remaining lifetime, while the MPC out of labor income depends on the length of the earning span relative to the total lifespan.

Both the PIH and the LCH, and their modern reformulation via the Euler equation, provide a coherent explanation of the three stylized facts. Cross-section data show a falling APC and an MPC substantially below one because transitory income variation dominates permanent variation across households at a point in time. Short-run time-series data (business cycle frequencies) yield an MPC out of current income that is lower than the long-run average, because temporary fluctuations in income are largely offset by smoothing behavior. Long-run time-series data display a stable, near-constant APC, because secular income growth reflects permanent movements in productivity and resources to which consumption adjusts proportionally.

The LCH carries profound implications for macroeconomic analysis and policy. Because consumption depends on net worth, changes in asset prices (stock market booms, housing bubbles) can have powerful effects on aggregate demand through wealth-induced changes in consumption. A temporary tax cut raises lifetime wealth by only the amount of the cut, so the induced increase in consumption is small. In contrast, a permanent tax cut triggers a near-unitary MPC out of the permanent income change. Hence, temporary fiscal stimulus has a low multiplier relative to permanent measures. Because saving rates vary systematically with age, the aggregate saving rate depends on the age structure of the population. Economies with a large share of prime-age workers will tend to have high saving rates, while rapidly ageing societies will experience declining aggregate saving as retirees dissave. The LCH forms the analytical basis for evaluating social security systems. A pay-as-you-go pension reduces the need for private retirement saving, thereby lowering national saving unless offset by government surpluses.

Extensions of the basic life-cycle model relax many of its simplifying assumptions. Introducing uncertainty about lifespan and income leads to precautionary saving and a consumption path that is not perfectly smooth but tracks buffer-stock behavior. Liquidity constraints can prevent young agents from borrowing against high future income, tilting the consumption profile more toward current earnings and explaining excessive sensitivity to predictable income changes. Bequest motives modify the terminal condition to allow positive assets at death, altering the timing of saving but preserving the basic logic of intertemporal allocation. Despite these complications, the core contribution of Modigliani and Brumberg endures: the recognition that consumption is determined by lifetime resources, that saving serves to transfer purchasing power from high-income to low-income periods, and that aggregate consumption behavior cannot be understood without accounting for the demographic structure and wealth distribution of the population.