The Role of Fiscal Policy in Britain s Great In ation in the 1970s

Transcription

1 The Role of Fiscal Policy in Britain s Great In ation in the 1970s Jingwen Fan 1 Patrick Minford 2, 3 Zhirong Ou 2 1 Nottingham Trent University 2 Cardi University 3 CEPR Fan, Minford & Ou () 1 / 20

2 1. UK Macro Environment and Policy during the 1970s. High in ation over this period (double-digit): an average of 14% (annual); peak of 36% captured in UK data ( ; un ltered, seasonally adjusted) Nominal int. rates Inflation Primary deficit Output y* Fan, Minford & Ou () 2 / 20

3 Monetary policy given no target before Fiscal policy was expansionary. During this period the UK economy seemed to have been governed by: non-ricardian sical policy; accommodative monetary policy; These broadly meet the policy environment requried by the Fiscal Theory of the Price level (FTPL). Fan, Minford & Ou () 3 / 20

4 2. The Fiscal Theory of the Price Level - what does it say? FTPL: Unbounded scal policy that just requires the real value of current liabilities to be met by the present value of future primary surpluses. Future price level (and hence expected in ation) is expected: a unique price level required given the outstanding debt. Financial markets set interest rates recursively to ensure the bond prices re ect in ation. Money supply follows to ensure the above are permitted. Thus, active scal policy accommodated by passive monetary policy (Leeper, 1991). Fan, Minford & Ou () 4 / 20

5 3. Identi cation between the FTPL and Orthodox Models Identi cation problem pointed by Cochrane (1999, 2001, and 2005): Equations that appear to re ect the FTPL model could have been generated by an Orthodox model; and vice versa. The two kinds of model can be confused with each other because no formal di erence in reduced form. Way out: These over-identi ed models are nevertheless numerically distinguishable. (Cross-equation restrictions and rational expectations impose di erent numerical restrictions in detail 4.) 4 In the paper, we show with Monte Carlo experiement that these models are clearly identi ed: when the FTPL is assumed true, it rejects the Orthodox model 24.5% of the time at the signi cance level of 5%. When the Orthodox model is true, the FTPL model is rejected 22.2% of the time. Fan, Minford & Ou () 5 / 20

6 4. An FTPL and an Orthodox Model for the UK in the 1970s Common components: y t y t = E t (y t+1 y t+1) 1 σ (Rs t E t π t+1 ) + err IS t (IS) π t = θ(y t y t ) + βe t π t+1 + err PP t (PP) y t y t 1 = c y + γ(y t 1 y t 2) + err y t (Equilibrium output) FTPL version: π t = κ(g t t t ) + c π + err pi t (FTPL) (g t t t ) = err g t t (Primary de cit) Fan, Minford & Ou () 6 / 20

7 Orthodox version: R s t = (1 ρ)[r ss + φ π π t + φ xgap (y t y t )] + ρr s t 1 + err R S t (Taylor Rule) 4(g t t t ) = δ[(g t) t 1 c g t ] + errt g t 0 (Primary de cit) Research question: can the models be the true (numerical) model generated the data? (...and which is more likely?) Fan, Minford & Ou () 7 / 20

8 5. The Method of Indirect Inference The Principle of Estimation/Testing using Indirect Inference Actual data Model(s) to be tested # (Bootstrap sim) Simulated data # # VECM representation VECM representation # # {z} vs Inference ( facts ) Wald statistic Distribution of inference (theory) Indirect, because it is the data s behavior rather than the data (likelihood) itself that is evaluated and compared. Fan, Minford & Ou () 8 / 20

9 A diagrammatic illustration Comparing the theory and the facts using Indirect Inference Auxiliary model: VECM(1) Data descriptors: VECM coe. and variance of the equ resid. (The gure here shows two desciptors along the horizontal axes.) Objective function: Wald = (Φ Φ) 0 1 (Φ Φ) (ΦΦ) Searching algorithm (Simulated Annealing) to test across all possible parameter values of structural model. Fan, Minford & Ou () 9 / 20

10 6.1 Model performance in replicating the VECM: joint-parameter Wald test Elements tested FTPL Orthodox Dynamics (VARX coe. only) 77.4 (.226) 78.1 (.219) Volatility (VARX resid. only) 6.9 (.931) 59.3 (.407) All elements (VARX coe.+resid.) 69.3 (.307) 84.9 (.151) Note: p-values in parentheses; p-value = (100-Wald percentile)/100. FTPL clearly better in volatility, but pretty much the same as Orthodox in dynamics. Both pass at the 95% con dence level overall, though FTPL is more probable. Fan, Minford & Ou () 10 / 20

11 6.2 A weighted in ation expectations model The two models cannot both be true because over-identi ed. One or both could be false, but simply not rejected because of low power. (But this can also mean that both models have some true components so that neither of them is strongly rejected) We investigate the possibility of both being false but some combination is true: viz such that in ation expectations are the probability-weighted outcome of FTPL and Orthodox. π t = W FTPL κ(g t t t ) +[1 W FTPL ] 1 1 [ φ π (1 ρ) (R t ρr t 1 ) φ xgap (y t yt ) α] +err Weighted π t Fan, Minford & Ou () 11 / 20

12 Performance of the Weighted model (p-values; as compared to the FTPL and the Orthodox) Elements tested FTPL Orthodox Weighted model Dynamics (VECM coe. only) Volatility (VECM resid. only) All elements (VECM coe.+resid.) Note: p-value = (100-Wald percentile)/100. The Weighted model brings clear improvement overall, mainly by improving the model dynamic behaviour. (Weight on FTPL (W FTPL ): 58% The best account of the period assumes in ation expectations were a probability-weighted combination of the FTPL and the Orthodox regimes. Fan, Minford & Ou () 12 / 20

13 6.3 Variance Decomposition Based on the Weighted Model Unit: % y π R s r err IS : err PP : err Weighted π : err y : err g t : Total: Output, norminal and real interest rates: mainly supply and in ation shocks. (the latter includes both monetary policy error and exogenous in ation error). In ation: dominated by scal shock, but supply shock is also important. Monetary shock did little. Thus the FTPL played a key role in causing the turbulent in ation. Fan, Minford & Ou () 13 / 20

14 7. What Happened to the UK Macroeconomy accoriding to the Weighted Model? - Output Q3 1974Q3 1975Q3 1976Q3 1977Q3 1978Q3 1979Q3 err(is) err(pp) err(weighted pi) err(y*) err(g t) Total Supply (red) and in ation (yellow) shocks dominated (but largely cancelled out each other). IS shock left being the main factor that caused the pre-1977 recession. Fan, Minford & Ou () 14 / 20

15 In ation 1973Q3 1974Q3 1975Q3 1976Q3 1977Q3 1978Q3 1979Q err(is) err(pp) err(weighted pi) err(g t) Total Fiscal (purple) and supply (red) shocks governed the in ation explosion from Others a ected little. Fan, Minford & Ou () 15 / 20

16 All but IS shock play a role; but mostly cancelled out each other. This explains why monetary policy played such a negligible role in setting in ation. Fan, Minford & Ou () 16 / 20 Nom. int. rates Q3 1974Q3 1975Q3 1976Q3 1977Q3 1978Q3 1979Q3 err(is) err(pp) err(weighted pi) err(g t) Total

17 Conclusion We found that: While the FTPL model explains the UK data behaviour better than the Orthodox model by a fair margin during the 1970s, the best accounts of the episode assumes in ation expectations is a probability-weighted average of the two; FTPL was operating but moderated by the Orthodox regime. The weighted model suggests that both poor expected scal and monetary policies caused the high in ation during this period. These, together with the weak productivity growth that deferred the recovery, seem to have paved the way for the decisive change both to the scal policy and to the monetary policies after Margaret Thatcher won her election in the late 1970s. Fan, Minford & Ou () 17 / 20

18 Appendix 1: Why Indirect Inference FAQ. Why Indirect Inference but not the now popular Bayesian? Indirect Inference as a method for model testing. Bayesian not (easily) applicable here because it requires true priors (which are hard to de ne in Macro). Test in the Baysesian context is done conditional on the priors (and so cannnot reject priors) which may well be wrong at its outset. Our purpose here is to test the models as a whole, including all their model structures and prior assumptions. We did attempt to estimate and compare the models using the Bayesian method. We found the ranking fragile as it can be easily overturned with priors in favor of one model against the other. Fan, Minford & Ou () 18 / 20

19 Why not FIML (that also tests the whole model)? FIML proved to have di culties allocating the optimal parameter values here with the small sample and at likelihood function of the models. (We see the Orthodox model has even slightly higher LR with the II estimates than with the FIML estimates. LR less powerful as a test compared to Indirect Inferece: II assesses a model s abilitly in replicating the data behaviour as re ected by the auxiliary model parameters; mis-speci ed model thus has mis-speci ed reduced form. LR assesses the a model s forecasting ability; a mis-speci ed model may still predict well. Fan, Minford & Ou () 19 / 20

20 Appendix 2: related literature of the FTPL Set out and developed by Leeper (1991), Sims (1994, 1997), Woodford (1996, 1998, 2001) and Cochrane (2001, 2005). Criticised by McCallum (2001, 2003) and Buiter (1999, 2002) on theoretical grounds of feasibility; however feasibility an empircial matter. Empirical assessments (our concern here) include Canzoneri, Cumby and Diba (2001), Bohn (1998), Woodford (1998b) and Davig et al. (2007); Loyo (2000), Tanner and Ramos (2003) - Brazil (late 1970s-early 1980s) Cochrane (1999) - US (1970s) Fan, Minford & Ou () 20 / 20