Common Errors in Endogenous Growth Analysis

Why Endogenous Growth Calculations Need an Assumption Check

Endogenous growth models connect choices such as research effort, education, or technology adoption to the path of productivity and output. Their conclusions can be sensitive to what is held constant and to how knowledge is produced. A calculation may be algebraically correct yet support the wrong conclusion if it confuses a productivity level with a growth rate, treats a private return as a social return, or applies a scale-effect result outside the model that produces it.

The diagnostic approach is to identify the model and its assumptions first, then state whether the change concerns a level or a rate, whose return is being measured, and which inputs or parameters remain fixed. These checks are aligned with the SOA VEE Economics (Validation by Educational Experience) syllabus.

Definition: A level effect changes the value of a variable relative to a counterfactual path, without necessarily changing its long-run growth rate. A growth-rate effect changes the percentage rate at which the variable grows over time. In continuous time, the growth rate of a positive variable $Z$ is $g_Z=\dot{Z}/Z$, where $\dot{Z}$ is its time derivative.

A one-time increase in productivity can put output on a permanently higher path while leaving its subsequent growth rate unchanged. That is different from a change that raises productivity growth every period. The distinction matters because the second change causes the gap between the two paths to widen over time.

Productivity Levels, Output Levels, and Growth Rates

Consider the Cobb–Douglas production function $Y=K^\alpha(AL)^{1-\alpha}$, with $0<\alpha<1$. Here $Y$ is output, $K$ is physical capital, $L$ is labor, and $A$ is labor-augmenting productivity. This production function assumes constant returns to scale in $K$ and effective labor $AL$. Holding $K$ and $L$ fixed, a one-time proportional increase in $A$ changes output by the elasticity $1-\alpha$, not by the full percentage increase in $A$.

\frac{Y_1}{Y_0} = \left(\frac{A_1}{A_0}\right)^{1-\alpha} \quad\text{when } K_1=K_0,\ L_1=L_0.

The formula follows by dividing the production function after the change by the function before it. With $K$ and $L$ unchanged, those factors cancel, leaving only the productivity ratio raised to the exponent $1-\alpha$. This is a finite-change result; it does not say that the percentage change in output equals the percentage change in productivity.

Worked Example: A One-Time Productivity Increase

A hypothetical economy has production $Y=K^{0.4}(AL)^{0.6}$. Its initial output is $\$1{,}000$ thousand. Productivity rises once by 5%, while capital and labor remain fixed. After that change, productivity follows its prior trend of 2% growth per year. Find the immediate output increase and the subsequent annual output growth rate. Round percentages to two decimal places.

Method. Apply the production function’s productivity elasticity for the immediate change, then calculate subsequent output growth using the same production function. The common mistake to avoid is treating the one-time 5% increase in the productivity level as a permanent 5-percentage-point increase in productivity’s annual growth rate.

The immediate productivity ratio is $A_1/A_0=1.05$. With $1-\alpha=0.6$, the output ratio is $\frac{Y_1}{Y_0}=1.05^{0.6}$. Using logarithms, $0.6\ln(1.05)=0.0292741\ldots$, so $1.05^{0.6}=\exp(0.0292741\ldots)\approx1.0297068$. The immediate percentage increase in output is therefore $(1.0297068-1)\times100\%\approx2.97068\%$, which rounds to 2.97%.

Output immediately after the shock is $\$1{,}000\times1.0297068=\$1{,}029.7068$ thousand, or $\$1{,}029.71$ thousand to the nearest cent. Once the one-time change has occurred, productivity grows at 2% annually. With capital and labor fixed, the annual output ratio is $1.02^{0.6}$. Since $0.6\ln(1.02)=0.0118816\ldots$, $1.02^{0.6}=\exp(0.0118816\ldots)\approx1.011952$. Subsequent output growth is about 1.1952%, or 1.20% per year to two decimal places.

Check. The initial shock raises output by less than 5%, as expected when the productivity elasticity is 0.6. The continuing annual growth rate comes from the 2% productivity trend, not from repeating the one-time 5% shock.

In this model, the growth-rate relationship can also be seen by taking the growth rate of each factor:

g_Y=\alpha g_K+(1-\alpha)(g_A+g_L).

This equation says that ongoing output growth depends on ongoing growth in capital, productivity, and labor. A one-time jump in $A$ is not itself a lasting increase in $g_A$. A common error is to insert the one-time percentage change in $A$ into the growth-rate equation as if that percentage recurred every year.

Private Returns Are Not Necessarily Social Returns

Research can create value for the firm that performs it and also generate knowledge that benefits other firms. The firm’s private return counts benefits it can capture. The social return counts the private benefit plus external benefits to others, net of the resources used. If spillovers are positive and are excluded from the firm’s calculation, private marginal benefit is below social marginal benefit at a given research effort.

The comparison must use marginal quantities: the extra private or social benefit from one more unit of effort, compared with the extra cost of that unit. A private optimum satisfies private marginal benefit equal to marginal cost. A social optimum instead equates social marginal benefit and marginal cost. The two choices need not coincide.

Worked Example: Research Effort With a Knowledge Spillover

A hypothetical research program chooses effort $n\geq0$, measured in standardized units per year. Its private benefit is $B_P(n)=12n-n^2$, its benefit to other firms is $B_E(n)=4n$, and its resource cost is $C(n)=4n+0.5n^2$. Find the private and socially optimal effort, and compare the net benefits at those choices. Treat the benefit and cost functions as present-value-equivalent units for this calculation.

Method. Differentiate the relevant benefit and cost functions, then equate marginal benefit to marginal cost. The common mistake to avoid is using the private benefit function alone to claim that the private choice is socially optimal.

The private marginal benefit is $B'_P(n)=12-2n$, and marginal cost is $C'(n)=4+n$. The private condition is $12-2n=4+n$, so $8=3n$ and $n_P=8/3\approx2.67$ units per year.

Total social benefit includes both components: $B_P(n)+B_E(n)=16n-n^2$. Social marginal benefit is $16-2n$. The social condition is $16-2n=4+n$, so $12=3n$ and $n_S=4$ units per year.

At the private choice, private net benefit is $B_P(8/3)-C(8/3)$. The benefit is $12(8/3)-(8/3)^2=32-64/9=224/9$. The cost is $4(8/3)+0.5(8/3)^2=32/3+32/9=128/9$. Thus private net benefit is $96/9=32/3\approx10.67$ units.

At the social choice, total benefit is $16(4)-4^2=64-16=48$. Cost is $4(4)+0.5(4^2)=16+8=24$. Social net benefit is $48-24=24$ units. At the private choice, social net benefit is $B_P(8/3)+B_E(8/3)-C(8/3)=32/3+32/3-32/3=32/3\approx10.67$ units. The social choice therefore produces higher social net benefit in this model.

Check. At $n=4$, social marginal benefit is $16-2(4)=8$, equal to marginal cost $4+4=8$. The social choice is larger because the private calculation omits the positive external marginal benefit of 4 units.

This example is a diagnostic, not a universal numerical rule: if the external effect were negative, or if firms could capture all spillover benefits, the comparison would differ. The relevant question is always which benefits and costs the decision-maker includes.

Scale Effects Depend on the Knowledge-Production Assumption

A scale effect means that the size of the economy, often its labor force, affects the long-run productivity growth rate. It is not a conclusion that follows from the phrase “endogenous growth” alone. It follows from specific assumptions about how research labor creates ideas and how those ideas contribute to productivity.

For example, in a simple Romer-style specification, let $\dot A/A=\delta L_R$, where $L_R$ is the number of workers in research and $\delta>0$ is a productivity parameter. If a constant share $s_R$ of total labor $L$ is allocated to research, then $L_R=s_RL$, so $g_A=\delta s_RL$. Under those assumptions, a larger labor force at the same research share raises the predicted productivity growth rate. The result depends on the model’s linear relation between research labor and the growth rate.

A semi-endogenous specification can instead be written $\dot A=\delta L_R^\lambda A^\phi$, with $\delta>0$, $\lambda>0$, and $0<\phi<1$. If $L_R=s_RL$ and labor grows at constant rate $n$, then on a balanced growth path the productivity growth rate is constant. Dividing by $A$ gives $g_A=\delta L_R^\lambda A^{\phi-1}$. Taking the growth rate of this expression on the balanced growth path gives $0=\lambda n+(\phi-1)g_A$, hence $g_A=\lambda n/(1-\phi)$.

Under these assumptions, the balanced-growth rate depends on labor-force growth $n$, not on the initial size of the labor force. That contrasts with the simple scale-effect specification. It is an error to quote the simple model’s scale conclusion without checking which knowledge-production function is being used.

Worked Example: Testing for a Scale Effect

Compare two hypothetical economies. In the first, productivity follows $\dot A/A=\delta L_R$, with $\delta=0.01$ per researcher per year and a research share $s_R=0.10$. Its labor force is initially 100 workers. In the second, use the same parameters and research share, but the labor force is 200 workers. Calculate the implied annual productivity growth rates. Then consider a semi-endogenous model with $\lambda=0.5$, $\phi=0.75$, and labor-force growth $n=0.02$ per year; calculate its balanced-growth productivity rate and state whether doubling the initial labor-force size changes that rate.

Method. Substitute research labor into each model’s stated growth relationship. The common mistake to avoid is treating a scale-effect result from one specification as if it also followed from the other.

In the first economy, $L_R=0.10(100)=10$ researchers. Therefore $g_A=0.01(10)=0.10$, or 10% per year in the model’s continuous-time rate units. In the second, $L_R=0.10(200)=20$, so $g_A=0.01(20)=0.20$, or 20% per year. The simple model predicts a scale effect: doubling labor at a fixed research share doubles the productivity growth rate.

For the semi-endogenous model, $g_A=\lambda n/(1-\phi)$. Substitution gives $g_A=0.5(0.02)/(1-0.75)=0.01/0.25=0.04$, or 4% per year. Doubling the initial labor-force size while keeping its growth rate at 2% does not change this balanced-growth rate. It may affect transition paths and levels, but the stated formula for the balanced-growth rate depends on $n$, $\lambda$, and $\phi$, not on the initial labor-force level.

Check. The first model explicitly makes $g_A$ proportional to the number of researchers. The second model’s balanced-growth condition eliminates the level of labor and leaves $g_A=\lambda n/(1-\phi)$.

A Practical Diagnostic Sequence

When a growth calculation seems to imply a large or surprising conclusion, check it in a fixed order. This separates arithmetic errors from errors about the model’s meaning.

Name the variable that changes.
Is it productivity $A$, productivity growth $g_A$, research effort, labor-force size, or a parameter of the knowledge-production function?

Classify the change as a level or rate change.
A one-time rise in $A$ changes its level. A permanent change in $g_A$ changes the slope of the path of $\ln A$.

Identify whose return is counted.
Include external knowledge benefits for a social-return calculation; do not assume a firm captures benefits received by others.

List the scale assumptions.
Check how research labor enters idea production, whether the research share is fixed, and whether the result concerns a transition or a balanced-growth rate.

Check units and the horizon.
Distinguish annual percentage growth from a one-time percentage change, and use the stated time units for rates and labor growth.

Common Errors and Their Corrections

Wrong: “Productivity rises by 5%, so output grows 5% every year from now on.” Correction: A one-time rise changes the productivity level; future output growth depends on subsequent growth in productivity, capital, and labor.
Wrong: “The firm’s marginal return to research equals the social marginal return.” Correction: This is true only when external benefits or costs are absent or fully internalized.
Wrong: “A larger population always raises long-run productivity growth.” Correction: That conclusion depends on the knowledge-production specification and on what is held fixed.
Wrong: “A growth rate is a level.” Correction: A growth rate describes percentage change per unit of time; a level is the amount of the variable at a particular time.
Wrong: “A model’s balanced-growth result describes every short-run response.” Correction: A balanced-growth rate is a long-run property; changes in levels and transition paths may differ.

Key takeaway: Diagnose an endogenous-growth claim by stating the model’s assumptions, separating levels from growth rates, and identifying whether the calculation is private or social. Scale effects are model-dependent, not a universal property of growth theory.

Check Your Understanding

For each problem, identify the relevant model distinction before calculating.

A hypothetical economy has $Y=K^{0.3}(AL)^{0.7}$. With $K$ and $L$ fixed, productivity rises once by 8%. What is the immediate percentage increase in output? Round to two decimal places. (Answer: $1.08^{0.7}-1\approx5.54\%$.)
In a simple model, $g_A=0.006L_R$, where $L_R$ is researchers. A country has 150 workers and allocates 8% to research; later its labor force is 225 with the same research share. Find the two productivity growth rates and their difference. (Answer: $L_R=12$ and $18$, giving $7.2\%$ and $10.8\%$, an increase of $3.6$ percentage points.)
A research project has private benefit $B_P(n)=10n-n^2$, external benefit $B_E(n)=2n$, and cost $C(n)=2n+0.5n^2$. Find private and social optimal effort for $n\geq0$. (Answer: Private $n=8/3$; social $n=10/3$.)
In a semi-endogenous model, $\lambda=0.4$, $\phi=0.8$, and labor-force growth is $1.5\%$ per year. Find balanced-growth productivity growth and explain the effect of doubling the initial labor force while leaving its growth rate unchanged. (Answer: $g_A=0.4(0.015)/(1-0.8)=3\%$ per year; doubling initial labor size does not change the balanced-growth rate.)
A firm reports that its research yields a private marginal benefit of 9 units and costs 7 units at the current effort. The research also creates an external marginal benefit of 3 units. Is the current marginal social net benefit positive, and by how much? (Answer: Yes; social marginal benefit is $9+3=12$, so marginal social net benefit is $12-7=5$ units.)

What You'll Learn