LH Advanced Macroeconomics

(07 33109)

Part 3: The Microfoundations of

Consumption & Investment

Dr Ceri Davies
[email protected]

Introduction
• We know that household-level consumption choices and
firm-level investment choices are important at the macro
level.

• Society divides its resources between consumption (for the

present) and investment (for the future); the latter is
particularly important for long-run living standards.

• For short-run fluctuations: we know that consumption and

investment are key elements on the ‘demand-side’ of the
economy; the demand for goods.

• We need to develop our understanding of C and I from the

household/firm perspective, i.e. in a micro-founded way.

1
 UK Data: Y, C, I

Source: Haskel (2019)

I: Consumption

• Consumption under Certainty: The Permanent-

Income Hypothesis

• Consumption under Uncertainty: The Random-Walk

Hypothesis

• Empirical Application: Testing the Random-Walk

Hypothesis

• Evaluating the consumption Euler equation

• Extensions + alternatives

2
 Consumption under Certainty: The PIH
• Consider an individual who lives for T periods; his/her
lifetime utility is (assume discount rate = 0):

𝑇
𝑢′ • > 0
𝑈 = ෍ 𝑢(𝐶𝑡 )
𝑡=1 𝑢′′ • < 0

• Allow this individual to have initial wealth A0 and labour

income of {Y1, Y2, Y3… YT} over the T periods of life.

• Saving/borrowing takes place at the market interest rate;

assume that r is exogenous and set = 0 for simplicity. The
individual is not permitted to die with outstanding debt.

The Budget Constraint

𝑇 𝑇

෍ 𝐶𝑡 ≤ 𝐴0 + ෍ 𝑌𝑡
𝑡=1 𝑡=1

• Since the M.U. of consumption is always positive, the

individual will not wish to die with any unused resources.

• Therefore, LHS<RHS would not be rational and LHS>RHS is

not permitted; we are left with only one option for the
budget constraint:
𝑇 𝑇

෍ 𝐶𝑡 = 𝐴0 + ෍ 𝑌𝑡
𝑡=1 𝑡=1

3
 The Constrained Maximisation Problem
𝑇 𝑇 𝑇

𝐿 = ෍ 𝑢 𝐶𝑡 + 𝜆 𝐴0 + ෍ 𝑌𝑡 − ෍ 𝐶𝑡
𝑡=1 𝑡=1 𝑡=1

• The individual chooses Ct; the FOC at time t is:

𝑢′ 𝐶𝑡 − 𝜆 = 0

• This holds in every time period. In general, for any period t+j:

The marginal utility of consumption

𝑢′ 𝐶𝑡+𝑗 − 𝜆 = 0
is constant in this case

Result for Ct
• Because of the assumptions we make about the utility
function, each level of consumption maps to a unique
M.U. of consumption.

• The result on the previous slide therefore implies a

constant level of consumption: C1 = C2 = C3… = CT
• Use this fact to write the budget constraint as:

𝑇 𝑇
1
𝑇𝐶𝑡 = 𝐴0 + ෍ 𝑌𝑡 𝐶𝑡 = 𝐴0 + ෍ 𝑌𝑡
𝑇
𝑡=1 𝑡=1

Divide lifetime resources equally across periods of life.

4
 Interpretation
• Friedman’s argument: consumption at any given point in
time depends upon lifetime income, not on income
earned in that particular period (Yt).

• Under the assumptions used here, we find a constant Ct:

𝑇
1
𝐶𝑡 = 𝐴0 + ෍ 𝑌𝑡
𝑇
𝑡=1

Permanent income

• In Friedman’s terminology, observed income at any point in

time can be split into Permanent income and Transitory
income; the former is much more important than the latter.

Implication for Savings

• An unexpected windfall gain (X) in period t is unlikely to
change permanent income by very much (X/T); a relevant
example could be a temporary tax cut.

• However, transitory income of this type can affect the

savings rate. Use the budget constraint to write:

𝑇
1 1
𝑆𝑡 = 𝑌𝑡 − 𝐶𝑡 = 𝑌𝑡 − ෍ 𝑌𝑡 − 𝐴0
𝑇 𝑇
𝑡=1

• Result: saving is high when income is high relative to its

‘long-run’ level, i.e. when transitory income is high;
individuals use saving (or borrowing) to smooth C.

5
 The PIH: Conclusions + Next Steps
• Friedman’s theory can be used to explain the
‘consumption puzzle’ which arises from the Keynesian
consumption function; Kuznets’ challenge – the ratio
of consumption to income is remarkably stable in time
series data.

• This has important implications for policymakers, e.g.

if they attempt to enact ‘stabilisation policy’ in the
form of a temporary tax cut.

• Robert Hall effectively extends the PIH to allow for

uncertainty; thus a role for expectations.

Consumption Under Uncertainty:

The Random Walk Hypothesis
• Suppose the individual faces uncertainty about his/her
(labour market) income in each period.

• Retain the assumption that the real interest rate and the
discount rate = 0; add rational expectations now.

• Assume that instantaneous utility is quadratic:

𝑇
𝑎
𝐸[𝑈] = 𝐸 ෍ 𝐶𝑡 − 𝐶𝑡2 (𝑎 > 0)
2
𝑡=1
𝑇 𝑇

Use the same budget constraint as before: ෍ 𝐶𝑡 ≤ 𝐴0 + ෍ 𝑌𝑡

𝑡=1 𝑡=1

6
 Individual Behaviour
• Consider the following scenario:
– Suppose that the individual has chosen an optimal
‘consumption plan’ for current and future periods.
– Consider a reduction in period 1 consumption (C1) of ΔC
and an equal increase in consumption at some future
date (t).
– If the individual is continuously optimising, this switch in
C should not affect overall utility; familiar logic by now.

Utility cost Utility benefit

0 = −(1 − 𝑎𝐶1 )Δ𝐶 + 𝐸1 [1 − 𝑎𝐶𝑡 ]Δ𝐶

So we can write:

𝐸1 1 − 𝑎𝐶𝑡 = 1 − 𝑎𝐶1 for 𝑡 = 2, 3, … , 𝑇

Certain terms can be removed from the expectations

operator. We can write the previous expression as:

1 − 𝑎𝐸1 𝐶𝑡 = 1 − 𝑎𝐶1 for 𝑡 = 2, 3, … , 𝑇

Therefore: 𝐸1 𝐶𝑡 = 𝐶1 for 𝑡 = 2, 3, … , 𝑇

The individual expects to consume the same amount as today

(C1) in any subsequent time period. This expectation is based
on the information set E1; new information (‘news’) could
arrive but this cannot be known ex ante.

7
 Explaining this Result

Recall: 𝐸1 𝐶𝑡 = 𝐶1 (i)
for 𝑡 = 2, 3, … , 𝑇

For example: 𝐸1 𝐶2 = 𝐶1
You could
iterate this
Roll forward: 𝐸2 𝐶3 = 𝐶2 process
again and
Substitute for C2: 𝐸1 𝐸2 𝐶3 = 𝐶1 again up to
time T…
Using the law of
𝐸1 𝐶3 = 𝐶1
iterated expectations: …this
explains (i)
In general: 𝐸1 𝐶𝑡 = 𝐶1

The budget constraint will hold exactly (for the same reasons
as before); individuals know this so, in expectation, we have:

𝑇 𝑇

෍ 𝐸1 𝐶𝑡 = 𝐴0 + ෍ 𝐸1 [𝑌𝑡 ]
𝑡=1 𝑡=1

The result E1[Ct]=C1 implies that the term on the LHS is:
𝑇

෍ 𝐸1 𝐶𝑡 = 𝑇𝐶1
𝑡=1

𝑇
1 Consume a fraction
Therefore
𝐶1 = 𝐴 + ෍ 𝐸1 [𝑌𝑡 ] of lifetime resources
for the B.C 𝑇 0 in period 1
𝑡=1

8
 Plan for the Remainder of the Topic

• Present the Random Walk Hypothesis and assess

its empirical validity.

• Evaluate the (consumption) Euler equation

empirically; remember, this was used to derive
the NKIS in the previous topic.

• Extensions + Alternatives.

• Investment: covering both theoretical and

empirical aspects once again.

The Random Walk Hypothesis

Given our findings so far, the difference between actual
consumption in period t+1 and what it was expected to be
one period beforehand (period t) can be expressed as:

𝐶𝑡+1 − 𝐸𝑡 𝐶𝑡+1 = 𝑒𝑡+1

where et+1 is a variable such that: 𝐸𝑡 𝑒𝑡+1 = 0

From our solution, we can write the expression above as:

𝐶𝑡+1 − 𝐶𝑡 = 𝑒𝑡+1 𝐶𝑡 = 𝐶𝑡−1 + 𝑒𝑡

Result: changes in consumption are unpredictable;
consumption itself follows a random walk (Hall, 1978).

9
 The RW Hypothesis: Intuition
• If consumption is expected to change it suggests that the
individual could do a better job of smoothing consumption
over his/her life (T years, say).
• Imagine that consumption is expected to increase between
today and tomorrow; this implies that the M.U. of
consumption today exceeds the (expected) M.U. of
consumption tomorrow [since U’’(C)<0].
• Optimal response: increase consumption today, reduce
consumption tomorrow; stop adjusting C when the marginal
utilities are equal, which implies no expected change in
consumption and Etet+1=0 (same argument for Et-1et=0).
• Only ‘news’ can change consumption plans when the
individual is already optimising; this cannot be predicted.

Testing the Random Walk Hypothesis

• Hall’s result was novel at the time; a major departure from
previous models of consumption, e.g. the Keynesian
consumption function.
• The traditional account of a recession was that when output
falls, consumption falls but is expected to recover; this implies
predictable changes in consumption which are not compatible
with Hall’s model.
• We need a decline in permanent income to generate a drop in
consumption in Hall’s model, an extension of Friedman’s PIH;
so downturns in income are expected to be permanent(?!)
• The idea that consumption responds to predictable
movements in income – thus refuting Hall’s theory – is known
as excess sensitivity (Flavin, 1981).

10
 Empirical Findings
• Hall (1978): regresses the change in consumption on variables
that are known in period t-1; should find statistically
insignificant estimates according to the RW Hypothesis; Hall’s
estimates confirm this for US data.

• Campbell and Mankiw (1989): use an instrumental variables

approach to address criticisms of Hall’s econometric test. Their
findings indicate substantial departures from the predictions of
the RW Hypothesis in US data. They conclude that ‘permanent
income’ is still an important variable for consumption, it’s just
not the whole story.

• Hsieh (2003): uses micro-level data. Finds that consumption

does not respond to predictable changes in income for the
Alaska Permanent Fund, thus supporting the PIH-RWH view.

Evaluating the Consumption Euler Equation

• Consider the paper by Canzoneri et al. (2007).

• Recall from the previous topic (but allow for uncertainty

here):

𝑈 ′ (𝐶𝑡 ) = 𝛽(1 + 𝑟𝑡 )𝐸𝑡 𝑈′ 𝐶𝑡+1

𝑃𝑡
Use the Fisher equation: 1 + 𝑟𝑡 = (1 + 𝑖𝑡 )𝐸𝑡
𝑃𝑡+1

𝑃𝑡
𝑈 ′ (𝐶𝑡 ) = 𝛽(1 + 𝑖𝑡 )𝐸𝑡 𝑈′ 𝐶𝑡+1
𝑃𝑡+1

11
• We can write as:

1 𝑈′ 𝐶𝑡+1 𝑃𝑡
= 𝛽𝐸𝑡
1 + 𝑖𝑡 𝑈 ′ (𝐶𝑡 ) 𝑃𝑡+1

• This is important because it provides a direct link between

monetary policy (𝑖𝑡 ) and household consumption choices
– and ultimately ‘aggregate demand – in many different
macro models, e.g. the ‘canonical New Keynesian model’.

• Canzoneri et al. (2007) ask how well the nominal interest

rate implied by the consumption Euler equation above fits
the observed nominal interest rate for US time-series data
(1966-2004).

For ‘Standard’ (CRRA) Preferences

𝐶𝑡1−𝜃 Coefficient of relative risk

𝑢(𝐶𝑡 ) = aversion, 𝜃>0
1−𝜃

Implies a nominal interest rate path of:

−𝜃
1 𝐶𝑡+1 𝑃𝑡
= 𝛽𝐸𝑡
1 + 𝑖𝑡 𝐶𝑡 𝑃𝑡+1

Calibration: 𝜃=2 𝛽=0.993

How does this compare to US data since the 1960s? Plots

that follow are expressed in terms of rmodel vs. ex-post rdata

12
 Source: Canzoneri et al. (2007, p.1867)

Corr(data, model) = -0.37 for the real interest rate; improves

for the nominal interest rate but still poor (0.20).

Particularly poor performance during key economic periods,

e.g. the Volcker disinflation (1980s) and the 2001 recession.

How to Proceed?
• Suggestion: modify the functional form for the
instantaneous utility function; perhaps the CRRA form is
not appropriate.

• We know from the previous topic that the NKIS

relationship can produce behaviour which is ‘too forward-
looking’ with ‘standard’ preferences; consumption plans
can change too quickly in response to a shock, for
instance.

• ‘Consumption habits’ will slow the dynamics of

consumption down (e.g. Dennis, 2009); there are several
different ways to model this.

13
 For Example: Fuhrer (2000)
Suggests the following functional form for utility (0≤γ≤1):

1−𝜃 Where the ‘habit’ element is:

1 𝐶𝑡
𝑢(𝐶𝑡 , 𝑍𝑡 ) =
1 − 𝜃 𝑍𝑡𝛾 𝑍𝑡 = 𝜌𝑍 𝑍𝑡−1 + (1 − 𝜌𝑍 ) 𝐶𝑡−1

The parameter 𝛾 shows the importance of the habit.

Fuhrer (2000) finds that ρZ≈0 empirically, so let’s set this = 0.

We have:
1−𝜃
1 𝐶𝑡
𝑢 𝐶𝑡 , 𝑍𝑡 = 𝛾
1 − 𝜃 𝐶𝑡−1

Implication: utility is no longer separable over time because

the consumption choices made today affect utility in the
next period.

Rewrite the utility function as:

1−𝜃
1 𝐶𝑡 1−𝛾
𝑢 𝐶𝑡 , 𝑍𝑡 = 𝐶
1 − 𝜃 𝐶𝑡−1 𝑡−1

You can now see that utility today depends upon both the
level of consumption and the growth in C from last period.

Households will now want to smooth the level of

consumption, as usual, but also its growth rate; this is one
way to model consumption habits but there are other ways.

14
 Source: Canzoneri et al. (2007, p.1869)

Corr(data, model) = -0.07 for the real interest rate; slightly

worse for the nominal interest rate (-0.10).

Large variations in r are required to overcome the strength

of the habit; we get a better fit for consumption but the
price is a rmodel which is much more variable than the data.

Internal vs. External Habit

• Fuhrer’s (2000) specification is an example of ‘internal

habit formation’. I benchmark my utility today against
my own utility in the past.

• Alternatively, with ‘external habit formation’ I

benchmark my utility today against some aggregate
level of consumption; e.g. ‘keeping up with the Jones’’,
Abel (1990).

• Canzoneri et al. (2007) consider different forms of

‘internal’ and ‘external’ habit formation; each model
generates a different Euler equation and, therefore, a
different expression for 𝑖𝑡 .

15
 Summary of Empirical Results
Different models of
consumption habits

Source: Canzoneri et al. (2007, p.1867)

Overall: this presetns a challenge for many macro models.

Extensions + Alternatives
• Hsieh’s (2003) findings in support of the PIH view possibly stem
from the large magnitude of Alaska Permanent Fund payments;
many other studies reject the PIH-RW view of consumption
behaviour (see Romer, 2019, pp.385-86 for discussion).
• We could allow for financial markets and liquidity constraints;
return to this in the next topic.
• We could allow for incomplete information, e.g. ‘sticky
information’ (Mankiw and Reis, 2002).
• We could drop rational expectations in favour of an alternative,
e.g. ‘learning’ (Evans and Honkapohja, 2008).
• We could drop constrained optimisation in favour of
approaches found in the Behavioural Economics literature, e.g.
bounded rationality ‘satisficing’.

16
 II: Investment

• Investment and the Cost of Capital.

• A Model of Investment with Adjustment Costs.

• Tobin’s q.

• Empirical Application: Tobin’s q and Investment.

• Extension with Uncertainty.

Our Starting Point

• Consider a firm that can rent capital at the price rK.

• The firms profits at any given point in time are given by (in
nominal terms):

𝑅 𝐾, 𝑋1 , … , 𝑋𝑛 − 𝑟𝐾 𝐾

Quantity of All other factors Notation for the first

capital chosen (taken as given) and second derivative:

𝑅𝐾 > 0 𝑅𝐾𝐾 < 0

• First order condition:
Rent capital until MRP of
𝑅𝐾 𝐾, 𝑋1 , … , 𝑋𝑛 = 𝑟𝐾 capital equals its rental price

17
 The User Cost of Capital
• In reality, most capital is not rented but owned by the firm
that is using it.

• Consider a firm which owns a unit of capital; assume that

the real market price of capital at time t is pK(t); think about
the decision to retain or dispose of (sell) that unit of capital.

• Retaining the capital has three costs:

i. The interest forgone from disposing of (i.e. selling) the
capital and saving the proceeds.
ii. Capital depreciation.
iii. Any change in the market price of capital, i.e. the disposal
value.

• This gives us a user cost of capital of (in real terms):

𝑟𝐾 𝑡 = 𝑟 𝑡 𝑝𝐾 𝑡 + 𝛿𝑝𝐾 𝑡 − 𝑝ሶ𝐾 (𝑡)

= 𝑟 𝑡 + 𝛿 − 𝑝𝑝ሶ 𝐾(𝑡)
𝑡
𝑝𝐾 𝑡
𝐾

• Consider favourable tax treatment for business investment,

e.g. an investment tax credit (i.e. a tax deduction), where 𝑓
is the amount of expenditure subject to the tax scheme and
𝜏 is the marginal tax rate.
• This will reduce the user cost of capital as follows:

𝑟𝐾 𝑡 = 𝑟 𝑡 + 𝛿 − 𝑝𝑝ሶ 𝐾 𝑡
𝑡
(1 − 𝑓𝜏)𝑝𝐾 𝑡
𝐾

and increase firms’ demand for investment accordingly.

18
 Two Major Shortcomings
1. A discrete change in an exogenous variable (e.g. r) leads
to a discrete change in the desired capital stock; but the
rate of change of capital is given by investment minus
depreciation, so a discrete increase in K, say, requires an
infinite increase in investment.

2. There is no forward-looking element to the model and

thus no role for expectations; surely firms attempt to
anticipate future economic conditions when deciding to
proceed with an investment project or not?

To fix (1) we introduce adjustment costs; then

extend to the case with uncertainty to address (2).

A Model with Capital Adjustment Costs

• We model ‘internal’ capital adjustment costs, e.g. the costs of
installing new machines or training workers how to use them.

• There are a large number (N) of identical firms in the

economy; each firm is therefore ‘small’.

• A representative firm’s real profits at time t are proportional

to its own level of capital stock, k(t), but decreasing in the
aggregate level of capital stock, K(t). More capital generates
more revenues and more costs at the firm level (e.g. constant
returns to scale) but at the aggregate level a rise in K
increases the marginal product and price of other factor
inputs; this reduces profit for the firm for each level of capital
input. Factor markets become more ‘crowded’.

19
 Convex Adjustment Costs
• This is the key part of the model.

• The rate of change of an individual firm’s capital stock is

defined as 𝑘,ሶ which in general could be <0, >0 or =0.

ሶ satisfy:
• The adjustment costs, denoted 𝐶(𝑘),

𝐶 0 =0 The firm faces a cost to either

increasing or decreasing its capital
𝐶′ 0 = 0 stock; the marginal adjustment cost
is increasing in the size of the
𝐶 ′′ (•) > 0 adjustment

• We also assume that there are no external adjustment

costs, e.g. the process of acquiring or disposing of capital
can itself change the unit price of capital. Remember, we
have many identical (small) firms.

• The price of a unit of capital is always 1 (purchase price

and disposal price).

• For simplicity, set the depreciation rate to 0. Then:

The rate of change of capital

𝑘ሶ 𝑡 = 𝐼(𝑡) equals the investment flow

• Firms maximise profits as follows (assume constant r):

∞
Π = න 𝑒 −𝑟𝑡 𝜋 𝐾 𝑡 𝑘 𝑡 − 𝐼 𝑡 − 𝐶 𝐼 𝑡 𝑑𝑡
𝑡=0

20
 Discrete Time Counterpart
∞
1
෩=෍
Π 𝜋(𝐾𝑡 )𝑘𝑡 − 𝐼𝑡 − 𝐶 𝐼𝑡
(1 + 𝑟)𝑡
𝑡=0

Capital accumulation eq’n when

depreciation = 0; there are infinitely
The Lagrangian problem is:
many constraints over time

∞ ∞
1
𝐿=෍ 𝜋(𝐾𝑡 )𝑘𝑡 − 𝐼𝑡 − 𝐶 𝐼𝑡 + ෍ 𝜆𝑡 (𝑘𝑡−1 + 𝐼𝑡 − 𝑘𝑡 )
(1 + 𝑟)𝑡
𝑡=0 𝑡=0

In this case, the Lagrangian multiplier (𝜆𝑡 ) can be interpreted as

the marginal impact of an exogenous increase in kt on the
lifetime value of the firm’s profits, discounted back to period t=0.

Rewrite the Lagrangian problem as:

∞
1
𝐿′ =෍ 𝜋(𝐾𝑡 )𝑘𝑡 − 𝐼𝑡 − 𝐶 𝐼𝑡 + 𝑞𝑡 (𝑘𝑡−1 + 𝐼𝑡 − 𝑘𝑡 )
(1 + 𝑟)𝑡
𝑡=0

where: 𝑞𝑡 ≡ (1 + 𝑟)𝑡 𝜆𝑡

Take the first order condition for 𝐼𝑡 :

1
−1 − 𝐶 ′ 𝐼𝑡 + 𝑞𝑡 = 0
(1 + 𝑟)𝑡

Which can be written as: 1 + 𝐶 ′ 𝐼𝑡 = 𝑞𝑡

21
 Interpretation
The equilibrium condition is:

1 + 𝐶 ′ 𝐼𝑡 = 𝑞𝑡

Purchase price of The value of the

a unit of capital Marginal capital to the firm
adjustment cost

The firm invests in capital stock until the cost = the benefit.

The difference in this model compared to previous

specifications is that the firm needs to take the adjustment
costs into account now.

The FOC for kt

• The firm will have some desired level of k in mind so we
can also take the first order condition for the level of
capital employed:

1 1
𝜋(𝐾𝑡 ) − 𝑞𝑡 + 𝑞 =0
(1 + 𝑟)𝑡 (1 + 𝑟)𝑡+1 𝑡+1

We need to this part because kt-1

enters the constraint; lead these
terms forward one time period

Then: 1 + 𝑟 𝜋 𝐾𝑡 = 1 + 𝑟 𝑞𝑡 − 𝑞𝑡+1

22
 Define: Δ𝑞𝑡 ≡ 𝑞𝑡+1 − 𝑞𝑡

Then we can write:

1
𝜋 𝐾𝑡 = (𝑟𝑞𝑡 − Δ𝑞𝑡 )
1+𝑟

Marginal revenue The opportunity cost of a unit of capital:

product of capital forgone interest mitigated by capital gains

This equilibrium condition is equivalent to the condition that

the firm rents capital to the point where its marginal revenue
product equals its rental price in the model without
adjustment costs.

Interpretation of q
• The variable q turns out to be a sufficient statistic to
capture all future information relevant to the firm’s
investment decision today.

• Specifically, q shows how an additional dollar of capital

affects the present value of the firm’s profits.

• The firm will want to increase its capital stock if q is high;

reduce it if q is low.

• Reminder: qt is the present discounted value of the

future marginal revenue products of a unit of capital:

𝑞𝑡 ≡ (1 + 𝑟)𝑡 𝜆𝑡

23
 Further Interpretation
• A unit increase in the firm’s capital stock increases the
present value of the firm’s profits by q, raising the market
value of the firm by the same amount.

• Take two firms (A,B) which are identical in every way except
for the fact that Firm A has one more unit of capital than
Firm B; Firm A’s market value is higher by q.

• Since we assume that the purchase price of one unit of

capital is fixed at 1, q/1 also shows the market value of one
unit of capital relative to its cost of acquisition.

• Recall: 1 + C’(It) = qt. Increase k if the market value of

capital exceeds its cost of acquisition; decrease k in the
opposite case.

Tobin’s q
• The ratio of the market value to the purchase price of
capital is known as Tobin’s q (Tobin, 1969).

• The purchase price of capital is sometimes called the

replacement cost of capital.

• In our constrained maximisation problem, we have

marginal q. This is the ratio of the market value of a
marginal unit of capital to its replacement cost.

• However, in practice we often need to use average q due

to measurement issues (e.g. tax considerations); this is
the ratio of the total value of the firm to the replacement
cost of its total capital stock.

24
 Phase Diagram Analysis
• Phase diagrams are often used to analyse dynamic
systems.

• We focus on two variables, K (aggregate) and q; take one

as given, we can then determine the other.

• For K, given q: recall, there are N identical firms in the

model; they face the same q and therefore choose the
same value for I.

• The rate of change of the capital stock (𝐾𝑡ሶ ) will be given by

the number of firms multiplied by some function of q:

∆𝐾𝑡 ≡ 𝐾𝑡ሶ = 𝑁𝑓(𝑞𝑡 )

Recall the equilibrium condition for one firm:

1 + 𝐶 ′ 𝐼𝑡 = 𝑞𝑡

Assuming that C’(It) is such (𝑞𝑡 − 1)

𝐼𝑡 =
that we can recover It 𝐶′

𝑁(𝑞𝑡 − 1)
Then: ∆𝐾𝑡 ≡ 𝐾𝑡ሶ = 𝑁𝑓 𝑞𝑡 =
𝐶′

So 𝑓(𝑞) is increasing in q, where we assumed that 𝐶 ′ (𝐼)

is increasing in 𝐼 (i.e. denominator >0) and 𝐶′ 0 = 0
Result: 𝐾ሶ < 0 when q<1; 𝐾ሶ > 0 when q>1; 𝐾ሶ = 0 when
q=1

25
 The Dynamics of K

If q>1, firms wish to increase

their capital stock

K constant if q=1

If q<1, firms wish to

decrease their capital stock

Source: Romer (2019, p.433)

For q, given K
Recall:
1
𝜋 𝐾𝑡 = (𝑟𝑞𝑡 − Δ𝑞𝑡 )
1+𝑟

Δ𝑞𝑡 ≡ 𝑞𝑡ሶ = 𝑟𝑞𝑡 − (1 + 𝑟)𝜋 𝐾𝑡

When 𝑞𝑡ሶ = 0 (without time subscripts now):

𝑟𝑞 = (1 + 𝑟)𝜋 𝐾 𝑞 = (1 + 𝑟)𝜋 𝐾 /𝑟

26
 The Dynamics of q

𝑞 = (1 + 𝑟)𝜋 𝐾 /𝑟

Downward sloping because

π(K) is decreasing in K

Also:

𝑞𝑡ሶ = 𝑟𝑞𝑡 − (1 + 𝑟)𝜋 𝐾𝑡

Source: Romer (2019, p.433)

The Full Dynamic System

We can now
identify a ‘stable’
saddle path, i.e. a
unique equilibrium
at point E

Source: Romer (2019, p.434)

27
 Empirical Findings
• Summers (1981) analyses US data (1931-1978); he finds a weak
relationship between investment (as a proportion of the total
capital stock) and q. His estimates imply very high capital
adjustment costs; it then takes a decade for the capital stock to
adjust just half of the way to its new equilibrium value following a
shock, which seems implausible. Blundell et al. (1992) for the UK.

• However, empirical studies face the challenge that ‘marginal q’,

which appears in the theory, is extremely difficult to measure in
actual data (e.g. we would need marginal tax rates for each firm).

• Cummins, Hassett, and Hubbard (1994) use several US tax

reforms as a ‘natural experiment’. Their estimates are more
plausible than Summers’ and suggest that q could be a relevant
indicator of the incentive to invest in K.

Extension: Allowing for Uncertainty

• Just like when we added uncertainty to the PIH model,

we can add uncertainty to our model of investment.

1
• Recall: 𝜋 𝐾𝑡 = (𝑟𝑞𝑡 − Δ𝑞𝑡 )
1+𝑟

• Rearrange as before and simply add the expectations

operator where appropriate:

𝐸𝑡 [Δ𝑞𝑡 ] ≡ 𝐸𝑡 [𝑞𝑡ሶ ] = 𝑟𝑞𝑡 − (1 + 𝑟)𝜋 𝐾𝑡

Our equilibrium condition now holds in expectation.

28
 Summary/Conclusions
• We have studied Consumption and Investment using the
‘micro-founded’ approach for macroeconomics developed in
this part of the module.

• For consumption: we have seen that our models may still be

too simple to capture key features of the data in a holistic (i.e.
general equilibrium) way.

• For investment: introducing capital adjustment costs is an

important step but the empirical performance of such models
is still mixed.

• For both: one (intentional) oversight so far is the role of

financing and financial markets more generally; this leads us
into the next topic.

29