Explain the 2-stage least squares (2SLS) method of estimation for the following simultaneous equation systems :
Cₜ = α + βYₜ + uₜ
Yₜ = Cₜ + Zₜ
where
C = Aggregate consumption expenditure
Y = National income
Z = Non-consumption expenditure
uₜ = A stochastic variable
10 marks · directive: Explain
The core of a good answer
Because Yₜ depends on uₜ through the identity, OLS on the consumption function is biased and inconsistent; 2SLS first regresses Yₜ on the exogenous Zₜ to get Ŷₜ = π̂₀ + π̂₁Zₜ (purged of uₜ), then regresses Cₜ on Ŷₜ, which here gives β̂ = Σcₜzₜ/Σyₜzₜ (deviations from means) and α̂ = C̄ − β̂Ȳ, a consistent estimator identical to indirect least squares and instrumental variables because the consumption equation is exactly identified.
An outline written by upsc.help, not an official UPSC answer; every fact checked by a separate review. How we work
What the question asks
Explain why OLS fails (simultaneity bias), check identification, derive the reduced form, then describe the two stages for this specific Keynesian model and give the resulting estimator and its properties.
Answer outline
Introduction
Endogenous: Cₜ, Yₜ. Exogenous: Zₜ (non-consumption expenditure, mainly investment), assumed uncorrelated with uₜ. uₜ has mean 0, constant variance σ², no autocorrelation. The second equation is an identity.
Step 1: Simultaneity bias of OLS
- Substituting: Yₜ = α/(1 − β) + Zₜ/(1 − β) + uₜ/(1 − β).
- Cov(Yₜ, uₜ) = σ²/(1 − β) ≠ 0, so the regressor is correlated with the error.
- plim β̂_OLS = β + (1 − β)σ²/(σ_Z² + σ²) > β: OLS overstates the marginal propensity to consume.
Step 2: Identification and reduced form
- Consumption function excludes Z (one predetermined variable excluded) and has one right-hand endogenous variable: K − k = 1 = m − 1, exactly identified.
- Reduced form: Yₜ = π₀ + π₁Zₜ + vₜ and Cₜ = π₂ + π₃Zₜ + vₜ with π₁ = 1/(1 − β), π₃ = β/(1 − β), vₜ = uₜ/(1 − β).
Step 3: Stage 1
- Regress Yₜ on a constant and Zₜ by OLS: Ŷₜ = π̂₀ + π̂₁Zₜ, with π̂₁ = Σyₜzₜ/Σzₜ² (lower case: deviations from means).
- Yₜ = Ŷₜ + v̂ₜ; Ŷₜ is a linear function of the exogenous variable only, so asymptotically uncorrelated with uₜ.
Step 4: Stage 2
- Regress Cₜ on a constant and Ŷₜ: Cₜ = α + βŶₜ + (uₜ + βv̂ₜ).
- β̂_2SLS = Σcₜŷₜ/Σŷₜ² = Σcₜzₜ/Σyₜzₜ, since ŷₜ = π̂₁zₜ.
- α̂ = C̄ − β̂Ȳ.
- This equals the IV estimator with Zₜ as instrument and the ILS estimator π̂₃/π̂₁.
Step 5: Properties and practice
- Consistent but biased in small samples; asymptotically normal.
- Standard errors must use residuals Cₜ − α̂ − β̂Yₜ (actual Y), not those of the stage 2 regression; packages do this automatically.
- For over-identified equations 2SLS gives a unique estimate where ILS gives several.
Conclusion
2SLS replaces the endogenous income variable by its fitted value from the reduced form, removing its correlation with the disturbance; for this exactly identified consumption function it yields β̂ = Σcz/Σyz, a consistent estimate of the marginal propensity to consume.
Examples and data to cite
- 2SLS was developed independently by Theil (1953) and Basmann (1957).
Facts to remember
- Simultaneity: Cov(Yₜ, uₜ) ≠ 0 makes OLS inconsistent.
- Order condition K − k ≥ m − 1; equality means exact identification.
- 2SLS = OLS of y on X̂, X̂ = Z(Z′Z)⁻¹Z′X.
- In an exactly identified equation 2SLS = ILS = IV.