Take the two builds to their nominal conclusions. India at a real 4.83% with the central bank's 4% inflation target gives nominal 9.018%. The United States at a real 2.25% with 2% inflation gives nominal 4.296%.
Should the terminal growth rate match the risk-free rate?
Now set those beside the risk-free rates that sit inside your cost-of-capital stacks. The United States long-run nominal growth of 4.296% sits 0.10 points above an illustrative 10-year Treasury yield of 4.20% (not a dated quote; check the current level on the Treasury's daily par yield curve). The correspondence is not luck: a currency's long-run nominal risk-free rate is roughly expected inflation plus an expected real rate, and over long horizons the real rate and real growth track each other, which is precisely why the terminal-growth cap can be written as the risk-free rate and still mean something.
India shows the other case, and it is the more instructive one. Nominal long-run growth of 9.018% sits 4.52 points above a cleaned INR risk-free rate of about 4.50% (check against the sovereign curve and the current default-spread deduction). Those numbers do not contradict each other, because they describe different dates. The 9.018% describes India over the coming decade. The terminal value in a rupee DCF describes India in the 2040s and beyond, by which time the capital-output ratio has risen, the dividend window has closed, the labour term has turned, and growth has fallen toward whatever the residual supports. The cap binds hard in India for exactly that reason, and an analyst who lifts today's nominal growth into a terminal value has assumed the transition never happens.
Advertisement
What does one point of terminal growth cost?
What the error costs is easy to size. At a WACC of 11.0% on a terminal cash flow of 100, terminal value at g = 4.5% is 1607.69 and at g = 5.5% it is 1918.18. One point of terminal growth is 19.31% of terminal value, and in most DCFs terminal value is well over half of enterprise value.
So the discipline is short and it closes the loop from macro back to a valuation you have already built. Write the country growth build. Take its mature-state conclusion, not its next-decade conclusion. Check it against the risk-free rate in your own stack, in the same currency. If your terminal growth exceeds that number, the model is asserting that one company outgrows its economy for ever, and the compounding arithmetic from the opening of this node says what that means.
{
"skill": "Running and reading a cross-country convergence regression",
"use": "Have an assistant assemble a panel from a primary cross-country source and run the initial-income regression on the full sample and on a fundamentals-matched subsample, reporting the slope, its standard error and the implied convergence speed for each, so that the absolute-versus-conditional distinction is a computed result rather than an assertion.",
"tools": ["Chat assistants with code execution - Claude, ChatGPT, Gemini", "Python: pandas plus statsmodels for the regression and its standard errors", "Penn World Table and World Bank World Development Indicators as the primary panels"],
"prompt": "Here is a panel with columns country, initial-year real GDP per worker at PPP, and average annual growth of real GDP per worker over the following T years. Write Python that (a) regresses growth on the natural log of initial GDP per worker, reporting slope, standard error, t statistic and R-squared, (b) repeats the regression on a stated subsample, (c) converts each slope b into an implied convergence speed beta = -ln(1 + b*T)/T and a half-life ln(2)/beta, flagging any slope for which 1 + b*T is not positive, and (d) prints the group means of both variables for each subsample. State T explicitly and do not round intermediate values.",
"verify": "Reproduce the slope by hand on the group means before trusting the regression, because a sign error here inverts the entire conclusion. Check that the subsample was chosen on fundamentals stated in advance rather than on the outcome, since selecting the countries that converged and then reporting convergence is the single most common defect in this literature and produces a beautiful t statistic. Trace initial income and growth to the same primary source and the same PPP base year, because a base-year revision moves these levels by more than the slope's standard error. Recompute the implied speed independently; Date-check the vintage of the panel; Cite the table and release (Guardrail 1-5). Report the full-sample and subsample results together or neither.",
"diy": "You must be able to state, without any code, what the two slopes mean, why a near-zero full-sample slope is not evidence against the Solow model, and how a regression slope becomes an annual convergence speed. Worked example 3 is the drill, and the mastery check asks you to convert a slope into a half-life by hand.",
"market": "US"
}