Ramsey-Cass-Koopmans Model

The problem

A representative household lives forever. It chooses a consumption path c(t) to maximize discounted lifetime utility:

max ∫0∞ e−ρt · u(c(t)) dt

where ρ > 0 is the rate of pure time preference (impatience), and u is a concave instantaneous utility function — typically CRRA: u(c) = c^1−σ / (1−σ). The household's resources accumulate according to the capital-accumulation equation, with k the per-capita capital stock, f(k) the production function (Cobb-Douglas, say), δ the depreciation rate, and n the population-growth rate:

k̇ = f(k) − (n + δ) · k − c

Output is split between consumption c and saving (which adds to capital after replacing depreciation and population dilution). The household picks the path c(t) optimally. This is a calculus-of-variations problem; Ramsey solved it in 1928 using Pontryagin's principle (avant la lettre). The solution emerges from two conditions: the Euler equation (intertemporal optimality) and the transversality condition (no perpetual borrowing).

The Euler equation

The first-order condition for intertemporal consumption choice in continuous time is:

ċ / c  =  (1/σ) · (f'(k) − δ − ρ − n·σ)

Or in discrete time, the more familiar form: u'(c_t) = β · (1+r_t+1) · u'(c_t+1), where β = 1/(1+ρ). Rearranged:

u'(ct) / u'(ct+1)  =  β · (1 + rt+1)

The intuition: at an optimum, the household is indifferent between consuming one unit today (marginal utility u'(c_t)) and saving it for tomorrow (giving up u'(c_t), gaining β·(1+r_t+1)·u'(c_t+1) in expected discounted utility). The Euler equation determines the growth rate of consumption from preferences (β, σ) and the return to capital (r).

When r > ρ, consumption grows; the household is saving in net. When r < ρ, consumption declines. The steady state has r = ρ (modified by depreciation and growth rates), and the consumption path is flat in per-capita terms.

Worked example — convergence to the modified golden rule

Calibrate: ρ = 0.04, σ = 1 (log utility), n = 0.01, δ = 0.05, capital share α = 1/3. The steady state requires f'(k*) = ρ + δ + n·σ = 0.04 + 0.05 + 0.01 = 0.10. With Cobb-Douglas, f'(k) = α·k^α−1, so:

α · k*α−1  =  0.10
(1/3) · k*−2/3  =  0.10
k*2/3  =  10/3  =  3.333
k*  =  3.3333/2  ≈  6.085

Same k* as Solow with s = 0.20! That's because at the Ramsey steady state, the implied savings rate works out to s = α(δ + n + ρ) / (δ + n + ρ + ρ/α). Plugging in: numerator = (1/3)(0.10) = 0.033; denominator = (0.10) + 0.04/(1/3) ≈ 0.10 + 0.12 = 0.22. s ≈ 0.15. The Ramsey saving rate is lower than the Solow golden-rule rate (which would be α = 0.33) because households discount the future and don't save the dynamically efficient maximum.

Now compare the golden rule (Solow) and modified golden rule (Ramsey). The pure golden rule maximizes steady-state consumption: f'(k_GR) = δ + n = 0.06. With α = 1/3: k_GR^2/3 = (1/3)/0.06 = 5.555, so k_GR ≈ 13.09. The Ramsey k* ≈ 6.09 is less than half of k_GR. Why? Because impatience ρ = 0.04 makes the household prefer present consumption to maximum steady-state consumption. The wedge ρ between the two steady states is the modified golden rule's defining feature.

The saddle path

Plot the dynamics in (k, c) space. Two key curves:

• The k̇ = 0 locus: c = f(k) − (n + δ)·k. This is a hump-shaped curve in (k, c) — capital accumulates below it, decumulates above it.
• The ċ = 0 locus: f'(k) = ρ + δ + n·σ. This is a vertical line at k = k*, the modified golden rule.

The two curves cross at the steady state (k*, c*). Around this point the dynamics are a saddle: trajectories with the wrong initial c either drive c to zero (consumption collapse) or send k to zero (capital depletion). Only one trajectory in each direction — the stable manifold — converges to the steady state. This is the saddle path.

The household's optimal policy: given initial k₀, choose c₀ exactly on the saddle path. The economy then converges along this path. This uniqueness of the initial choice is what gives the Ramsey model determinate consumption — unlike Solow, where any savings rate gives a different path, Ramsey's optimality picks one path.

Ramsey vs other growth models

	Ramsey (1928)	Solow (1956)	Overlapping Generations (Diamond 1965)	RBC (Kydland-Prescott 1982)	New Keynesian DSGE (Smets-Wouters 2003)	HANK (Kaplan-Moll-Violante 2018)
Saving	Optimal, ρ-discounted	Exogenous parameter s	Endogenous, finite-lived	Optimal + shocks	Optimal + nominal frictions	Optimal + heterogeneity
Agents	Single rep agent	Single rep agent	OLG cohorts	Single rep agent	Single rep household, firms	Continuum, heterogeneous
Steady state	Modified golden rule	s·f(k)=(δ+n)k	Possibly inefficient	Stochastic, ergodic	Stochastic + sticky prices	Stochastic + wealth dist.
Dynamics	Saddle path	Monotonic convergence	Period-by-period eq.	Stochastic Ramsey	Linearized + shocks	Partial nonlinear
Welfare	Pareto-optimal	Not from optimization	Can be Pareto-inefficient	Pareto-optimal	Sub-optimal (frictions)	Sub-optimal (markets incomplete)
Used in	Foundation of macro	Textbook growth	Public finance	RBC literature	Central banks	Modern frontier

Frank Ramsey — a brief famous-problem aside

Frank Plumpton Ramsey (1903–1930) was a 25-year-old Cambridge mathematician when he published "A Mathematical Theory of Saving" in the Economic Journal in 1928. He was also a foundational figure in probability, decision theory, and philosophy of language, and a close friend of Wittgenstein and Keynes. His central question: how much should a nation save? Ramsey set up the problem as constrained calculus of variations — long before Pontryagin formalized optimal control — and derived what is now called the Keynes-Ramsey rule:

u'(c) · ċ  =  (Bliss − u(c))

where Bliss is the maximum attainable utility. Ramsey assumed ρ = 0 (no discounting; "ethically indefensible") and the saving rate equates current marginal utility to the gap from Bliss. Cass and Koopmans rediscovered the framework in 1965 with positive discounting and the modern CRRA structure. Ramsey died of liver failure at age 26 — one of the most productive short careers in mathematics.

Counterarguments

Representative-agent fiction. Aggregating a heterogeneous population to a single optimizing household requires strong assumptions (Gorman aggregation: linear Engel curves) that fail empirically. Modern heterogeneous-agent models (HANK) retain the Euler equation at the household level but generate aggregate dynamics that pure Ramsey cannot.

Time-inconsistency. Exponential discounting at constant ρ is assumed. Hyperbolic discounting — empirically much closer to behavior — generates time-inconsistent plans (today's self wants to save more than tomorrow's self will actually do). Laibson (1997) and the behavioral literature build on this.

Liquidity constraints. Households cannot freely borrow against future labor income. The Ramsey first-order conditions assume frictionless asset markets; in practice many households are constrained. Krusell-Smith and related models handle this with substantial complications.

Transversality and bubbles. The transversality condition (lim k(t)·e^−ρt → 0) rules out perpetual debt accumulation, but selecting the saddle-path equilibrium implicitly assumes coordination. With incomplete markets or strategic interaction, multiple equilibria including bubbles can be consistent with all first-order conditions.

Common pitfalls

• Confusing the modified golden rule with the golden rule. The Ramsey steady state k* sits below the Solow golden rule k_GR because impatience drives a wedge.
• Treating the Euler equation as a consumption function. The Euler equation is an optimality condition, not a behavioural rule. It pins down the slope (growth) of consumption; the level still requires the transversality condition.
• Forgetting σ. The intertemporal elasticity of substitution 1/σ controls how willing the household is to substitute consumption across periods. Small σ → very willing → big response to interest rate changes; large σ → smooth path even with high r-ρ wedge.
• Reading "representative agent" too literally. The model captures aggregate dynamics; it doesn't predict that real households all hold the same consumption profile.
• Ignoring labor supply. Plain Ramsey treats labor as inelastically supplied at 1. Real Business Cycle and modern DSGE extensions add labor-leisure choice, materially changing comparative statics.