Health Economics

Math Review: Constrained Optimization

PPHA 383/PBPL 283/ECON 277, Spring 2024

Constrained Optimization: Equality Constraint

Start with a simple two-variable case. Let f (x1 , x2 ) be the objective function, with the
choice variable x1 , x2 satisfying the constraint g (x1 , x2 ) = 0. We can formally write the
problem as
maxx1 ,x2 f (x1 , x2 )
s.t. g (x1 , x2 ) = 0

A direct way to solve the problem is to substitute x2 for x1 . That is, we can find x2 as
a function of x1 from the constraint g (x1 , x2 ) = 0. The method is often useful when g
has a simple form such as linear. In this way, we rewrite the problem as

max f (x1 , x2 (x1 ))

x1

so that we are back to the world of optimization without constraint.

1
Constrained Optimization: Equality Constraint

Unfortunately, not all g are nice enough. That is why we want a systematic way to
treat constrained optimization problem. One such way is called Lagrangian, which is
done by building an auxiliary function L.
To be specific, if we have the original problem
maxx1 ,x2 f (x1 , x2 )
s.t. g (x1 , x2 ) = 0
Lagrangian is built by subtracting λg from f . The λ ∈ R is called the Lagrange
multiplier. The critical point of the problem is found by FOC.
L (x1 , x2 , λ) = f (x1 , x2 ) − λg (x1 , x2 )

From the math perspective, the value of λ doesn’t matter. Economists often formulate
it as the following, where λ is positive and can be interpreted as the shadow price.
2
L (x1 , x2 , λ) = f (x1 , x2 ) + λg (x1 , x2 )
Constrained Optimization: Lagrangian

Example: Consider the following problem

maxx1 ,x2 −ax12 − bx22
s.t. x1 + x2 − 1 = 0
Set the Lagrangian
L (x1 , x2 , λ) = −ax12 − bx22 + λ (x1 + x2 − 1)
∂L
= −2ax1 + λ = 0
∂x1
∂L
= −2bx2 + λ = 0
∂x2
∂L
= x1 + x2 − 1 = 0
∂λ
b a 2ab
It is easily solved that x1∗ = , x∗ = ,λ = .
a+b 2 a+b a+b 3
General Formulation of Utility Maximization

Consider the following problem

maxx,y U(x, y )
s.t. px x + py y ≤ I
Set the Lagrangian
L (x1 , x2 , λ) = U(x, y ) + λ (I − px x − py y )
∂L
= MUx − λpx = 0
∂x
∂L
= MUy − λpy = 0
∂y
∂L
= I − px x − py y = 0
∂λ

4
General Formulation of Utility Maximization

∂L
= MUx − λpx = 0
∂x
∂L
= MUy − λpy = 0
∂y
∂L
= I − px x − py y = 0
∂λ
From the first two equations, we have MUx = λpx and MUy = λpy . Take the ratio of
these two equations, we have
MUx λpx px
= =
MUy λpy py
Does the following sound familiar to you?
px
MRSxy =
py
5
Grossman Model: a simplified version with two periods

In class we talked about a simplified model with only 2 periods.

ˆ Ht = health in period t
ˆ Xt = all other goods you could buy in period t
ˆ δ = depreciation rate of health
ˆ r = interest rate
ˆ h = how much you invest in your health
ˆ ρ = time discount rate (the smaller this is, the more you value the future)

Problem set up:  

1
max U (X1 , H1 ) + U (X2 , H2 )
1+ρ
= (1 − 
s.t. H2  δ) (H1 + h)
1
Px X1 + Ph h + Px X2 = I
1+r 6
Grossman Model: a simplified version with two periods

Problem set up:  

1
max U (X1 , H1 ) + U (X2 , H2 )
1+ρ
= (1 − 
s.t. H2  δ) (H1 + h)
1
Px X1 + Ph h + Px X2 = I
1+r

The person wants to maximize the sum of their period utilities (where the second
period is reduced by their discount factor). The constraints: health in period 2
depends on investment and depreciation as well as their starting level of health, and
their income constraint.

7
Grossman Model: a simplified version with two periods

What are we maximizing over? — X1 , X2 , h

Plug in H2 definition and set up the Lagrangian:

  h  
i
1 1


max U (X1 , H1 ) + 1+ρ U X2 , (1 − δ) (H1 + h) + λ I − Px X1 + Ph h + 1+r Px X2

First-order conditions:
∂L
= UX1 − λ2 Px = 0
∂X1
   
∂L 1 1
= UX2 − λ2 Px = 0
∂X2 1+ρ 1+r
 
∂L 1
= Uh (1 − δ) − λPh = 0
∂h 1+ρ
 
∂L 1
= I − Px X1 − Ph h − Px X2 = 0
∂λ 1+r
8
Grossman Model: a simplified version with two periods

Some intuition from the FOC of h:

 
1
Uh (1 − δ) = λPh
1+ρ

ˆ Suppose people who don’t invest much in health, they will have a lower h.
ˆ Since the marginal utility is decreasing, they have a higher Uh (when h is low).
 
ˆ For the FOC of h to hold, we need either 1+ρ 1
or (1 − δ) to be smaller (or both).
ˆ This means that either ρ or δ needs to be larger (or both).
ˆ This translates into fast aging or value future less (or both).