Inter-Temporal Optimisation in Growth Models

Have you ever wondered why economists don’t just focus on making the best decision for today? The answer lies in understanding that economic choices aren’t isolated snapshots-they’re paths that unfold over time. When households save for retirement, governments invest in infrastructure, or firms accumulate capital, they’re engaged in inter-temporal optimization, making decisions that balance present costs against future benefits. This concept forms the backbone of modern economic growth theory, transforming our understanding from simple static trade-offs to dynamic pathways of economic development.

From static snapshots to dynamic paths 🔗

Static optimization is like taking a photograph-it captures the best choice at a single moment in time. You might use it to decide how much coffee to buy today given your budget, or how many workers a firm should hire this month. These problems ask: what is the optimal value right now?

Dynamic optimization, however, is more like filming a movie. It doesn’t just find the best point-it finds the best path over time. Instead of asking “what should I consume today?”, dynamic optimization asks “what consumption path should I follow over my entire lifetime?” This fundamental shift from points to paths is what makes growth models with optimizing agents so powerful for understanding economic development.

Think of planning a road trip. Static optimization would tell you the best speed to drive at this very moment. Dynamic optimization would map out your entire journey-when to speed up, when to slow down, where to stop for gas-to reach your destination in the best overall way.

The two-period consumption model: a simple starting point 🔗

To grasp inter-temporal choice, economists often begin with the simplest case: two periods, which we can call “today” and “tomorrow.” Imagine you earn income today and tomorrow, and you must decide how much to consume in each period.

Here’s where it gets interesting. You’re not stuck consuming exactly what you earn each period. If you expect to earn more tomorrow, you might borrow today to smooth out your consumption-after all, why live poorly today if you know riches are coming? Conversely, if you’re earning well today but expect leaner times ahead, you might save and lend at the prevailing interest rate.

Let’s make this concrete. Suppose you earn ₹50,000 today and expect ₹70,000 tomorrow, and the interest rate is 10%. You could consume all ₹50,000 today and all ₹70,000 tomorrow. Or you could borrow ₹10,000 today (consuming ₹60,000) and pay back ₹11,000 tomorrow (consuming ₹59,000). The interest rate creates a price for moving consumption across time-it tells you how many rupees tomorrow you must give up to have an extra rupee today.

The role of the interest rate 🔗

The interest rate acts as the linchpin in inter-temporal decisions. A higher interest rate makes future consumption cheaper relative to present consumption-it’s as if tomorrow’s goods are “on sale.” This encourages saving. A lower interest rate makes borrowing more attractive, encouraging consumption today. This price mechanism helps coordinate the plans of savers and borrowers across time, just as market prices coordinate buyers and sellers in space.

Deriving the inter-temporal budget constraint 🔗

The magic of the inter-temporal budget constraint is that it expresses all possible consumption combinations in present value terms. This gives us a single number that captures the total resources available across time.

Starting with the two-period case, if you earn income Y₁ today and Y₂ tomorrow, and the interest rate is r, your budget constraint in present value terms is:

C₁ + C₂/(1+r) = Y₁ + Y₂/(1+r)

The left side represents the present value of your consumption stream, and the right side represents the present value of your income stream. The constraint says these must be equal-you can’t consume more than you have, once everything is properly discounted to the present.

Notice what this means: if you want to consume more today (increase C₁), you must consume less tomorrow (decrease C₂), and the trade-off is governed by the interest rate. For every extra rupee consumed today, you give up (1+r) rupees of consumption tomorrow. This is the fundamental inter-temporal trade-off that households face.

Extending to many periods 🔗

This logic extends naturally to many periods. For a lifetime with T periods, the constraint becomes a sum of all discounted consumption equal to the sum of all discounted income. Modern intertemporal models use this framework to analyze everything from life-cycle savings behavior to optimal fiscal policy.

Understanding present value and discounting 🔗

Present value is the cornerstone concept that makes inter-temporal comparison possible. The fundamental question it answers is: what is a future payment worth to me today?

Suppose someone promises to pay you ₹10,000 five years from now. Is that as valuable as ₹10,000 today? Clearly not. Money today can be invested and earn returns, so you’d need less than ₹10,000 today to have ₹10,000 in five years. But there’s more to it than just financial returns-people are inherently impatient. Even without investment opportunities, most of us prefer gratification sooner rather than later.

This is where the discount factor beta (β) comes in. In economic models, β is a number between 0 and 1 that captures both the financial opportunity cost of time and our psychological time preference. A β of 0.95 means that a utility of 100 next year is worth only 95 units to you today. The lower your β, the more impatient you are.

Exponential discounting 🔗

The most common approach in economics is exponential discounting, where utility T periods in the future is discounted by β^T. This creates a consistent pattern: each additional period into the future reduces value by the same proportion. If β = 0.95, then something two years away is worth 0.95 × 0.95 = 0.9025 as much as the same thing today. This mathematical elegance makes exponential discounting tractable for modeling, though behavioral economists have identified situations where people deviate from this pattern.

What is a functional? 🔗

Here’s where things get mathematically deeper. In ordinary calculus, you learn about functions-rules that take a number and give you back another number. For instance, f(x) = x² takes 3 and gives you 9.

But in dynamic optimization, we need to work with something more sophisticated: functionals. A functional takes an entire function as its input and gives you back a number. Think of it as a “function of functions.”

Why do we need this? Because in inter-temporal problems, we’re not just choosing numbers (like how much to consume today). We’re choosing entire paths-functions that describe how variables like consumption or capital evolve over time. A functional evaluates these paths and assigns them a value.

An example from growth theory 🔗

In the famous Ramsey growth model, we want to maximize lifetime utility, which is the integral of discounted utility at each point in time. This is a functional: it takes the entire consumption path c(t) over time and spits out a single number representing total welfare. The mathematical form might look like the integral of e^(-ρt) × u(c(t)) from 0 to infinity, where ρ is the discount rate and u is the instantaneous utility function.

Finding the consumption path that maximizes this functional is the essence of inter-temporal optimization in growth models. We’re not choosing a consumption level-we’re choosing an entire consumption trajectory.

Methods for dynamic optimization 🔗

Economists have developed three powerful mathematical techniques for solving dynamic optimization problems, each with its own strengths and applications. Understanding when and how to use each method is crucial for analyzing growth models.

Calculus of variations 🔗

The oldest technique, dating back to the 18th century, is the calculus of variations. This method works beautifully when you want to find a path (a function) that minimizes or maximizes a functional. The key tool is the Euler equation, which provides necessary conditions for an optimal path.

The calculus of variations is particularly elegant for problems where the path itself is the object of choice, with well-defined starting and ending points. Classical applications include finding the shortest path between two points or the shape that minimizes surface area. In economics, it’s used when the entire trajectory from a starting state to a terminal state needs to be optimized.

Optimal control theory 🔗

Developed in the 1950s and popularized by Pontryagin, optimal control theory generalizes the calculus of variations by introducing control variables-decisions you can adjust at each point in time to influence the evolution of state variables. This is the method of choice for most modern growth models.

The central tool here is the Hamiltonian, which combines the objective function with the constraints on how state variables can evolve. The Hamiltonian’s first-order conditions give you both the equation of motion for the state variables and the co-state equations that describe the shadow prices of these variables. Optimal control theory has become the workhorse of dynamic economics, used in everything from resource extraction to monetary policy analysis.

Dynamic programming 🔗

The third approach, dynamic programming, flips the problem on its head. Instead of finding the entire optimal path from start to finish, you work backward. At each state, you ask: “If I’m here now, what’s the best immediate decision given that I’ll behave optimally in the future?”

This recursive structure is captured by the Bellman equation, which says that the value of being in a certain state equals the maximum of your immediate payoff plus the discounted value of the state you’ll reach next. Dynamic programming is especially powerful for problems with discrete time, uncertainty, or complex state spaces. It’s also the foundation of modern computational economics and reinforcement learning in artificial intelligence.

Which method should you use? 🔗

The choice depends on your problem’s structure. Calculus of variations works well for smooth, deterministic problems with fixed endpoints. Optimal control is ideal for continuous-time problems with state dynamics and no uncertainty. Dynamic programming excels when you have discrete decision points, stochastic elements, or need numerical solutions. Many researchers find that starting with optimal control theory provides the best intuition for economic growth models, though all three methods ultimately yield equivalent characterizations of optimal behavior.

What do you think? Have you ever made a decision that involved trading off benefits today against costs tomorrow-like choosing to invest in education or save for a major purchase? How did you weigh the present against the future, and looking back, do you think you got the inter-temporal trade-off right?

References

1. https://en.wikipedia.org/wiki/Ramsey–Cass–Koopmans_model
2. https://www.nber.org/papers/w0885
3. https://en.wikipedia.org/wiki/Discounted_utility
4. https://books.google.com/books/about/Dynamic_Optimization.html?id=0IoGUn8wjDQC