Dynamics of Real GDP Per Capita Growth

Transcription

1 Dynamics of Real GDP Per Capita Growth Daniel Neuhoff Humboldt-Universität zu Berlin CRC 649 June 2015

2 Introduction Research question: Do the dynamics of the univariate time series of per capita GDP differ across countries? Do previous results regarding the persistence hold up when accounting for model uncertainty using RJMCMC? Are there big differences in the impulse responses of GDP across economies internationally? What role does the detrending method play in this context? Aside: Sensitivity to time period? Aside: Which components of GDP drive the results for the US? My contribution: Use RJMCMC and account for model uncertainty Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 1/69

3 Preview of Results The dynamics of GDP differ across the the six countries studied Estimates of persistence in line with previous studies Ranking in terms of persistence robust to detrending device (except HP) Consumption seems to drive the results for the US The results are sensitive to the time period studied, at least for the UK Results from frequentist methods are mostly in line with results from RJMCMC. Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 2/69

4 Outline Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 3/69

5 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 4/69

6 Some Literature Trends and Random Walks: Nelson and Plosser (1982), Stock and Watson (1988),DeJong and Whiteman (1991), Perron and Wada (2009) Persistence: Campbell and Mankiw (1987), Campbell and Mankiw (1989), Cochrane (1988), Cheung and Lai (1992), Cheung (1994) Diebold and Rudebusch (1989), Koop (1991), Perron (1993) Persistence and models: Jones (1995), Ragacs and Zagler (2002), Fatas (2000b), Fatas (2000a), Steven N. Durlauf and Sims (1989) Time series and DSGE models: Wallis (1977) and Ravenna (2007) Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 5/69

7 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 6/69

8 A Canonical ARMA Model A non-stationary ARMA(p,q) model can be written as (1 L) D P(L)y t = y T t + Q(L)ε t where y t is the non-stationary observed data, D is the degree of integration, P(L) and Q(L) are autoregressive and moving average lag polynomials Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 7/69

9 Different Trends Here: 3 options for modeling trend (Log-) First differences (the (I) models): D=1, y T t = 0 t Linear trend fitted by OLS: D=0, y T t = a+γt HP-Filtering: D=0, y T t solves min{ yt T t= yt y T t 2+λ t= y T t+1 2yT t + yt t 1 2} Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 8/69

10 Stationary Model Zero mean autoregressive moving average process with orders p,q: y C t = P p 1 y t P p p y t p+ε t + Q q 1 ε t Q q q ε t q P p and Q q parameter vectors of the AR and MA polynomials ε t N(0,σ 2 ) Impose stationarity by reparametrizing the polynomials in terms of (inverse) partial autocorrelations For first differences de-mean before estimating Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 9/69

11 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 10/69

12 Overview Standard practice in DSGE estimation: Metropolis-Hastings samplers Now: Varying dimensionality of the parameter space Reversible Jump Markov Chain Monte Carlo Pioneered by Green (1995) as generalization of M-H samplers Allows for moves between parameter spaces of varying dimensionality Provides samples from a joint posterior distribution across different models and their corresponding parameter spaces Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 11/69

13 RJMCMC Algorithm 1. Set the initial stateς 0 of the Markov Chain 2. For i=1 to N 2.1 setς=ς i Propose a visit to model(p, q) with probabilityγ pq ((p, q) (p, q)) 2.3 Sample u fromγ u (ς, u) 2.4 Setς = g(ς, u) 2.5 Accept draw with probability α=min 1,χ(ς,ς ) with χ(ς,ς )= (ς ) (ς) }{{} Likelihood Ratio ρ(ς ) γ(ς ς ) ρ(ς) }{{} Prior Ratio γ(ς ς) g (ς, u) }{{} Proposal Ratio 2.6 If the draw is accepted setς i =ς. If the draw is rejected setς i =ς Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 12/69

14 Priors All a priori independent Object Prior p DU(0, 10) q DU(0, 10) (Inverse) Partial Autocorrelation U( 1, 1) σ ε IG(1,1) Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 13/69

15 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 14/69

16 Data Quarterly real GDP from OECD (VOBARSA) Population from OECD Quarterly real per capita GDP for 1960:1 : 2007:4 Canada, France, Italy, Japan, UK, US Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 15/69

17 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 16/69

18 The K-S Test Two-sample Kolmogorov-Smirnov test statistic given by KSS a,b = sup F a (x) F b (x) x The critical values for this statistic are given by n a + n b KSS α = c(α) n a n b where n a and n b are the sample sizes for empirical distributions a and b respectively and c(α) is a coefficient depending on the chosen significance level α: α c(α) K-S Test rejects null hypothesis for all posteriors of the persistence measures in all cases and combinations Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 17/69

19 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 18/69

20 Measures Let P(L) be the autoregressive polynomial and Q(L) the moving average polynomial P(L)y t = Q(L)ε t with infinite moving average representation y t = Q(L) P(L) ε t= C(L)ε t C(1) n : Sum of the first n parameters of the infinite moving average polynomial C(L) First differences: Sum of changes in log-gdp up to horizon n over and above the constant drift (Campbell and Mankiw (1987)) Trend-stationary: Same measure, giving cumulated, undiscounted departure from trend Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 19/69

21 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 20/69

22 Model Indicator Posteriors Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 21/69

23 Impulse Responses 1 1 US 1.2 UK 0.8 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Canada 1.2 France 0.8 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 22/69

24 Impulse Responses Italy 1.2 Japan Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 23/69

25 Observations The AIC and AICC criteria chose the same model for all countries respectively The model chosen by the BIC are close to the means and modes of the impulse responses from RJMCMC BIC and RJMCMC choose more parsimonious models Means and modes for the impulse responses from RJMCMC are very close Credible sets from RJMCMC impulse responses are tight Shape and persistence of the impulse responses differ Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 24/69

26 C(1) 1: "Normal" Impulse Response C(1) 20 US C(1) 40 US C(1) 60 US C(1) 20 Canada C(1) 40 Canada C(1) 60 Canada Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 25/69

27 C(1) 2: "Normal" Impulse Response C(1) 20 Italy C(1) 40 Italy C(1) 60 Italy C n (1) constant with increasing n Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 26/69

28 C(1) 3: Random Walk C(1) 20 UK C(1) 40 UK C(1) 60 UK C n (1)=1 as expected Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 27/69

29 C(1) 4: More Persistent C(1) 20 France C(1) 40 France C(1) 60 France C(1) 20 Japan C(1) 40 Japan C(1) 60 Japan C n (1) keeps growing with n Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 28/69

30 Point estimates: RJMCMC Horizon Canada 1.42[1.38] 1.45[1.39] 1.46[1.39] 1.46[1.39] 1.46[1.39] [1.16; 1.8] [1.16; 1.94] [1.16; 2] [1.16; 2.01] [1.16; 2.01] France 0.921[0.913] 1.07[1.05] 1.25[1.2] 1.43[1.3] 1.54[1.32] [0.726; 1.15] [0.744; 1.45] [0.746; 1.93] [0.746; 2.6] [0.746; 3.07] Italy 1.51[1.49] 1.55[1.5] 1.56[1.51] 1.58[1.51] 1.58[1.51] [1.22; 1.89] [1.22; 2.03] [1.22; 2.1] [1.22; 2.14] [1.22; 2.14] Japan 1.78[1.75] 2.33[2.28] 3.12[3.03] 4.11[3.92] 4.72[4.38] [1.42; 2.23] [1.77; 3.06] [2.2; 4.35] [2.39; 6.48] [2.4; 8.22] UK 1.01[1] 1.01[1] 1[1] 1[1] 1[1] [0.92; 1.12] [0.917; 1.13] [0.914; 1.13] [0.914; 1.13] [0.914; 1.13] US 1.56[1.54] 1.56[1.54] 1.52[1.54] 1.5[1.54] 1.5[1.54] [1.22; 1.98] [0.962; 2.11] [0.557; 2.14] [0.368; 2.14] [0.31; 2.14] Table 1. C n (1) for different horizons; RJMCMC estimates Clustering for Canada, France, Italy, and US Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 29/69

31 Point estimates: Frequentist Horizon Canada 1.68; ; ; ; ; 1.41 France 0.925; ; ; ; ; 1.31 Italy 1.32; ; ; ; ; 1.38 Japan 1.54; ; ; ; ; 5.1 UK 1.23; ; 1 1.2; ; ; 1 US 1.39; ; ; ; ; 1.44 Table 2. C n (1) for different horizons; frequentist estimates for AIC; BIC Mostly agree with estimates from RJMCMC, especially with BIC. Estimates contained in credible sets except for US w/ AIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 30/69

32 Comparison to Campbell and Mankiw (1989) Horizon: Canada [1.39] 1.46[1.39] 1.46[1.39] France [1.2] 1.43[1.3] 1.54[1.32] Italy [1.51] 1.58[1.51] 1.58[1.51] Japan [3.03] 4.11[3.92] 4.72[4.38] UK [1] 1[1] 1[1] US [1.54] 1.5[1.54] 1.5[1.54] Table 3. C n (1): Results from CM in the first row, posterior mean and median in the second Different behavior as horizon grows Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 31/69

33 Persistence Ranking Horizon Estimate Mn Md CM AIC BIC Mn Md CM AIC BIC Mn Md CM AIC BIC Canada France Italy Japan UK US Table 4. Ranking by persistence Ranking mostly coincides Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 32/69

34 Observations Persistence differs Shapes of posterior distributions differ There is some clustering Frequentist and RJMCMC estimates agree Different behavior of estimates compared to Campbell and Mankiw (1989), but ranking consistent Null hypothesis of same posterior distribution rejected for all cases at the 1% significance level Japan super-persistent, UK pure random walk Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 33/69

35 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 34/69

36 Model Indicator Posteriors Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 35/69

37 Impulse Responses US 1.2 UK Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Canada Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 36/69

38 Impulse Responses 2 2 Italy 4.5 Japan Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC France Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 37/69

39 C(1) 1: Least persistent C(1) 20 US C(1) 40 US C(1) 60 US C(1) 20 UK C(1) 40 UK C(1) 60 UK Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 38/69

40 C(1) 2: Medium persistence C(1) 20 France C(1) 40 France C(1) 60 France C(1) 20 Canada C(1) 40 Canada C(1) 60 Canada Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 39/69

41 C(1) 3: Most Persistence C(1) 20 Italy C(1) 40 Italy C(1) 60 Italy C(1) 20 Japan C(1) 40 Japan C(1) 60 Japan Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 40/69

42 Point Estimates RJMCMC Horizon Canada 7.62[7.55] 14[13.7] 24.7[24.1] 39.8[38.9] 50.1[48.2] [6.58; 8.9] [11.5; 17.3] [19.2; 31.9] [27; 55.8] [29.5; 76.9] France 5[4.96] 9.76[9.65] 19.8[19.3] 37.9[36] 50.8[47.6] [4.29; 5.81] [7.86; 12.1] [14.5; 26.8] [25.4; 57.2] [30.1; 82.4] Italy 8.06[8.02] 15.3[15.1] 28.9[28.4] 52.6[51.9] 72.7[72] [6.97; 9.28] [12.8; 18.4] [23.5; 35.8] [40.5; 67.1] [51.5; 96] Japan 8.62[8.6] 19.5[19.4] 45.5[44.7] 97.9[95] 139[133] [7.23; 10.1] [15.3; 24.4] [34.5; 59.5] [69; 137] [86.1; 212] UK 5.34[5.3] 8.64[8.55] 12.9[12.7] 17.1[16.1] 18.9[17.1] [4.77; 6.09] [7.15; 10.4] [9.39; 17.4] [10.2; 27.5] [10.3; 33.9] US 7.6[7.56] 12.4[12.2] 16.1[15.6] 17.8[16.3] 18.4[16.4] [6.55; 8.75] [9.88; 15.4] [10.7; 23.1] [10.1; 30.4] [10.2; 33.3] Table 5. C n (1) for different horizons; RJMCMC estimates Clustering: UK and US; France and Canada Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 41/69

43 Point Estimates Frequentist Horizon Canada 8.03; ; ; ; ; 45.3 France 4.4; ; ; ; ; 31.4 Italy 7.66; ; ; ; ; 24.7 Japan 7.45; ; ; ; ; 84.9 UK 6.11; ; ; ; ; 18.8 US 7.3; ; ; ; ; 11.7 Table 6. C n (1) for different horizons; frequentist estimates for AIC; BIC Behavior with changing n consistent with RJMCMC. Estimate for Italy significantly different. Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 42/69

44 Persistence Ranking Horizon Estimate Mean Median AIC BIC Mean Median AIC BIC Mean Median AIC BIC Canada France Italy Japan UK US Table 7. Ranking by persistence Ranking practically unchanged, except for US/UK. Despite greater differences in point estimates ranking mostly consistent with frequentist results Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 43/69

45 Observations Persistence ranking almost unchanged, with exception of US Clustering different Models less parsimonious with substantially higher persistence in impulse response BIC model choice not as close to RJMCMC as before Credible sets are wider Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 44/69

46 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 45/69

47 Model Indicator Posteriors Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 46/69

48 Impulse Responses US 1 UK Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Canada 1 France Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 47/69

49 Impulse Responses 2 1 Japan 1 Italy Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 48/69

50 Observations Estimates seem to be mostly driven by HP-filter artifacts (see King and Rebelo (1993) and Cogley and Nason (1995)) Use stochastic trends, fractional integration, or time trends Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 49/69

51 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 50/69

52 Major Components of US GDP Question: Which of the major components of US GDP seems to drive results? Analyze first-differenced real components of GDP: Private Consumption Government Consumpion Gross Fixed Capital Formation Imports Exports Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 51/69

53 Model Choice Figure 1. Posteriors for model indicators Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 52/69

54 Impulse Responses 1 4 Imports 4 Exports Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC 3 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 53/69

55 Impulse Responses Private Consumption 2 Gross Fixed Capital Formation Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC 1.5 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Government Consumption 1 US 0.8 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC 0.8 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Shape of impulse response function for US seems to be mostly driven by consumption Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 54/69

56 C(1): Consumption C(1) 20 Private Consumption C(1) 40 Private Consumption C(1) 60 Private Consumption C(1) 20 Government Consumption C(1) 40 Government Consumption C(1) 60 Government Consumption Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 55/69

57 C(1): Capital formation C(1) 20 Gross Fixed Capital Formation C(1) 40 Gross Fixed Capital Formation C(1) 60 Gross Fixed Capital Formation Bi-modal posterior Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 56/69

58 C(1): Imports, Exports C(1) 20 Imports C(1) 40 Imports C(1) 60 Imports C(1) 20 Exports C(1) 40 Exports C(1) 60 Exports Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 57/69

59 Observations Shape of impulse response appears to be mainly driven by consumption Negative response to shock from capital formation series Bi-modal posterior for C(1) for capital formation Imports and Exports random walk Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 58/69

60 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 59/69

61 Splitting the sample Question: Is the result for the UK driven by some large and persistence shifts? Is the finding robust? Split the sample in 1980 and 1990, the beginning and end of the reign of the Iron Lady Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 60/69

62 Model Choice Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 61/69

63 Impulse Responses :1-1979: :1-1989:1 1 Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC :1-2007: :1-2007: Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Mean Mode Posterior IRF 5% Bound Posterior IRF 95% Bound AIC AICC BIC Random walk only for subsamples starting in 1960! Significantly reduced standard error of disturbance for later subsamples Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 62/69

64 C(1): Split in 1980 C(1) :1-1979:4 C(1) :1-1979:4 C(1) :1-1979: C(1) :1-2007: C(1) :1-2007: C(1) :1-2007: Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 63/69

65 C(1): Split in 1990 C(1) :1-1989:4 C(1) :1-1989:4 C(1) :1-1989: C(1) :1-2007: C(1) :1-2007: C(1) :1-2007: Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 64/69

66 Persistence Horizon : [0.804] 0.76[0.804] 0.751[0.804] 0.749[0.804] 0.748[0.804] 1979:4 [0.348; 1] [0.208; 1] [0.139; 1] [0.117; 1] [0.114; 1] 1980:1-2.12[2.08] 2.36[2.21] 2.49[2.23] 2.56[2.23] 2.58[2.23] 2007:4 [1.54; 2.86] [1.53; 3.69] [1.5; 4.4] [1.5; 4.7] [1.5; 4.74] 1960: [1] 0.976[1] 0.975[1] 0.975[1] 0.975[1] 1989:4 [0.822; 1.06] [0.818; 1.06] [0.816; 1.06] [0.815; 1.06] [0.815; 1.06] 1990:1-2.35[2.28] 2.49[2.34] 2.54[2.34] 2.56[2.34] 2.56[2.34] 2007:4 [1.67; 3.27] [1.59; 3.91] [1.54; 4.22] [1.52; 4.29] [1.51; 4.3] Table 8. C n (1) for different horizons; RJMCMC estimates UK quite persistent for later subsamples, second only to Japan Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 65/69

67 Observations Random walk result not sensitive to time period studied After 1980, persistent response Substantially reduced standard deviation of shock Maggie? Luck? Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 66/69

68 Literature AR(I)MA Models Reversible Jump Markov Chain Monte Carlo Data Comparing Posteriors Persistence Measures Results First Differences Results Linear Trend Results HP-Filter US GDP Components UK Subsamples Conclusion Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 67/69

69 Conclusion Results are in line with previous work Differences in persistence and shape of impulse responses across countries Ranking in persistence seems to carry over from difference stationary to linear detrending HP-filter problematic Dynamics of US GDP appear to be mostly driven by consumption At least for the UK, results sensitive to time period First-differencing produces more parsimonious representations Different persistence needed for trend- and difference-stationary economic models Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 68/69

70 Thank you for your attention! Literature AR(I)MA RJMCMC Data Comparison Persistence FD LINEAR HP Components UK Conclusion 69/69

71 References I CAMPBELL, J., AND G. MANKIW (1989): International Evidence on the Persistence of Economic Fluctuations, Journal of Monetary Economics, 23, CAMPBELL, J. Y., AND N. G. MANKIW (1987): Are Output Fluctuations Transitory?, The Quarterly Journal of Economics, 102(4), CHEUNG, Y.-W. (1994): Aggregate output dynamics in the twentieth century, Economics Letters, 45, CHEUNG, Y.-W., AND K. S. LAI (1992): International evidence on output persistence from postwar data, Economic Letters, 38, COCHRANE, J. H. (1988): How Big Is the Random Walk in GNP?, Journal of Political Economy, 96(5), COGLEY, T., AND J. M. NASON (1995): Effects of the Hodrick-Prescott filter on trend and difference stationary time series - Implications for business cycle research, Journal of Economic Dynamics and Control, 19, DEJONG, D. N., AND C. H. WHITEMAN (1991): Reconsiderung Trends and random walks in macroeconomic time series, Journal of Monetary Economics, 28, DIEBOLD, F. X., AND G. D. RUDEBUSCH (1989): Long memory and persistence in aggregate output, Journal of Monetary Economics, 24(2), A-1/A-10

72 References II FATAS, A. (2000a): Do Business Cycles Cast Long Shadows? Short-Run Persistence and Economic Growth, Journal of Economic Growth, 5, (2000b): Endogenous growth and stochastic trends, Journal of Monetary Economics, 45, GREEN, P. J. (1995): Reversible Jump Markov Chain Monte Carlo, Biometrika, 82, JONES, C. I. (1995): Time Series Tests of Endogenous Growth Models, The Quarterly Journal of Economics, 110(2), KING, R. G., AND S. T. REBELO (1993): Low frequency filtering and real business cycles, Journal of Economic Dynamics and Control, 17, KOOP, G. (1991): Intertemporal Properties of Real Output: A Bayesian Analysis, Journal of Business & Economic Statistics, 9(3), NELSON, C., AND C. PLOSSER (1982): Trends and Random Walk in Macroeconomic Time Series, Journal of Monetary Economics, 10, PERRON, P. (1993): The HUMP-Shaped Behavior of Macroeconomic Fluctuations, Empirical Economics, 18, PERRON, P., AND T. WADA (2009): Let s take a break: Trends and cycles in US real GDP, Journal of Monetary Economics, 56, A-2/A-10

73 References III RAGACS, C., AND M. ZAGLER (2002): Persistence of Shocks to Output in Austria and Theories of Economic Growth, Empirica, 29, RAVENNA, F. (2007): Vector autoregressions and reduced form representations of DSGE models, Journal of Monetary Economics, 54, STEVEN N. DURLAUF, D. R., AND C. A. SIMS (1989): Output Persistence, Economic Structure, and the Choice of Stabilization Policy, Brookings Papers on Economic Activity, 2, STOCK, J. H., AND M. W. WATSON (1988): Variable Trends in Economic Time Series, The Journal of Economic Perspectives, 2(3), pp WALLIS, K. F. (1977): Multiple Time Series Analysis and the Final Form of Econometric Models, Econometrica, 45(6), A-3/A-10

74 Appendix: Information Criteria AIC=2k 2ln( ) AICC=AIC+ 2k(k+1) n k 1 BIC= 2ln( )+kln(n) with k being the number of model parameters and n the number of observations A-4/A-10

75 Appendix: Frequentist Regression Results First Differences Country Criterion P 1 P 2 P 3 P 4 P 5 Q 1 Q 2 Q 3 Q 4 Q 5 σe Canada AIC (0.150) (0.066) (0.053) (0.117) (0.161) (0.095) (0.097) (0.136) (0.063) AICC (0.150) (0.066) (0.053) (0.117) (0.161) (0.095) (0.097) (0.136) (0.063) BIC (0.065) (0.079) (0.057) (0.054) (0.057) (0.056) France AIC (0.108) (0.198) (0.117) (0.099) (0.083) (0.116) (0.155) (0.122) (0.097) AICC (0.108) (0.198) (0.117) (0.099) (0.083) (0.116) (0.155) (0.122) (0.097) BIC (0.091) (0.085) (0.038) (0.059) Italy AIC (0.212) (0.117) (0.225) (0.149) (0.127) (0.230) (0.181) (0.227) (0.203) (0.152) (0.058) AICC (0.212) (0.117) (0.225) (0.149) (0.127) (0.230) (0.181) (0.227) (0.203) (0.152) (0.058) BIC (0.057) (0.057) Japan AIC (0.031) (0.036) (0.036) (0.055) (0.034) (0.061) (0.095) (0.093) (0.088) (0.053) (0.084) AICC (0.031) (0.036) (0.036) (0.055) (0.034) (0.061) (0.095) (0.093) (0.088) (0.053) (0.084) BIC (0.032) (0.026) (0.025) (0.059) (0.031) (0.059) (0.090) UK AIC (0.095) (0.047) (0.092) (0.106) (0.073) (0.101) (0.059) (0.060) (0.071) AICC (0.095) (0.047) (0.092) (0.106) (0.073) (0.101) (0.059) (0.060) (0.071) BIC (0.054) US AIC (0.067) (0.049) (0.060) (0.043) (0.064) (0.068) (0.067) (0.052) (0.042) AICC (0.067) (0.049) (0.060) (0.043) (0.064) (0.068) (0.067) (0.052) (0.042) BIC (0.063) (0.047) A-5/A-10

76 Appendix: Frequentist Results Linear Detrending Country Criterion P 1 P 2 P 3 P 4 P 5 Q 1 Q 2 Q 3 Q 4 Q 5 σe Canada AIC (0.041) (0.033) (0.046) (0.079) (0.136) (0.136) (0.109) (0.059) (0.055) AICC (0.041) (0.033) (0.046) (0.079) (0.136) (0.136) (0.109) (0.059) (0.055) BIC (0.042) (0.033) (0.047) (0.073) (0.106) (0.062) (0.056) France AIC (0.037) (0.025) (0.043) (0.023) (0.029) (0.064) (0.051) (0.058) (0.081) AICC (0.037) (0.025) (0.043) (0.023) (0.029) (0.064) (0.051) (0.058) (0.081) BIC (0.066) (0.065) (0.057) (0.035) (0.059) Italy AIC (0.011) (0.014) (0.011) (0.010) (0.071) (0.129) (0.104) (0.094) (0.110) (0.059) AICC (0.011) (0.014) (0.011) (0.010) (0.071) (0.129) (0.104) (0.094) (0.110) (0.059) BIC (0.011) (0.014) (0.011) (0.010) (0.071) (0.129) (0.104) (0.094) (0.110) (0.059) Japan AIC (0.278) (0.117) (0.436) (0.057) (0.234) (0.272) (0.240) (0.219) (0.208) (0.087) (0.091) AICC (0.159) (0.107) (0.201) (0.056) (0.101) (0.136) (0.077) (0.101) (0.087) BIC (0.071) (0.108) (0.110) (0.078) (0.095) UK AIC (0.175) (0.079) (0.079) (0.148) (0.165) (0.263) (0.209) (0.078) (0.067) AICC (0.175) (0.079) (0.079) (0.148) (0.165) (0.263) (0.209) (0.078) (0.067) BIC (0.024) (0.051) US AIC (0.241) (0.133) (0.221) (0.056) (0.166) (0.264) (0.195) (0.118) (0.226) (0.043) AICC (0.241) (0.133) (0.221) (0.056) (0.166) (0.264) (0.195) (0.118) (0.226) (0.043) BIC (0.063) (0.059) (0.099) (0.041) A-6/A-10

77 Appendix: Frequentist Results US GDP Components Component Criterion P 1 P 2 P 3 P 4 P 5 Q 1 Q 2 Q 3 Q 4 Q 5 σe Exports AIC (0.048) (0.047) (0.070) (0.085) (0.094) (0.075) (0.087) (0.792) AICC (0.048) (0.047) (0.070) (0.085) (0.094) (0.075) (0.087) (0.792) BIC (0.003) (0.047) (0.080) (0.092) (0.081) (0.830) Government AIC Consumption (0.083) (0.037) (0.042) (0.036) (0.069) (0.109) (0.051) (0.056) (0.043) (0.111) (0.064) AICC (0.083) (0.037) (0.042) (0.036) (0.069) (0.109) (0.051) (0.056) (0.043) (0.111) (0.064) BIC (0.083) (0.037) (0.042) (0.036) (0.069) (0.109) (0.051) (0.056) (0.043) (0.111) (0.064) Gross Fixed AIC Capital Formation (0.022) (0.015) (0.077) (0.128) (0.123) (0.125) (0.088) (0.227) AICC (0.022) (0.015) (0.077) (0.128) (0.123) (0.125) (0.088) (0.227) BIC (0.068) (0.060) (0.216) Imports AIC (0.668) AICC (0.668) BIC (0.668) Private AIC Consumption (0.120) (0.074) (0.059) (0.081) (0.102) (0.061) (0.034) AICC (0.120) (0.074) (0.059) (0.081) (0.102) (0.061) (0.034) BIC (0.068) (0.064) (0.033) A-7/A-10

78 Appendix: Frequentist Results UK Subsamples Period Criterion P 1 P 2 P 3 P 4 Q 1 Q 2 Q 3 Q 4 Q 5 σe 1960:1-1979:4 AIC (0.336) (0.139) (0.286) (0.357) (0.281) (0.442) (0.164) (0.256) AICC (0.114) (0.165) BIC (0.181) 1980:1-2007:4 AIC (0.193) (0.124) (0.114) (0.144) (0.202) (0.126) (0.185) (0.167) (0.122) (0.031) AICC (0.193) (0.124) (0.114) (0.144) (0.202) (0.126) (0.185) (0.167) (0.122) (0.031) BIC (0.048) (0.104) (0.032) 1960:1-1989:1 AIC (0.139) (0.068) (0.135) (0.154) (0.110) (0.150) (0.086) (0.090) (0.150) AICC (0.139) (0.068) (0.135) (0.154) (0.110) (0.150) (0.086) (0.090) (0.150) BIC (0.113) 1990:1-2007:4 AIC (0.087) (0.094) (0.111) (0.089) (0.150) (0.136) (0.212) (0.166) (0.110) (0.022) AICC (0.087) (0.094) (0.111) (0.089) (0.150) (0.136) (0.212) (0.166) (0.110) (0.022) BIC (0.171) (0.304) (0.189) (0.075) (0.192) (0.323) (0.180) (0.017) A-8/A-10

79 K-S Tests 1 C(1) 5 Canada France Germany Italy Japan UK US Canada (*) (*) (*) (*) (*) (*) France (*) (*) 0.94 (*) (*) (*) (*) Germany (*) (*) (*) (*) (*) (*) Italy (*) 0.94 (*) (*) (*) (*) (*) Japan (*) (*) (*) (*) (*) (*) UK (*) (*) (*) (*) (*) (*) US (*) (*) (*) (*) (*) (*) 0 C(1) 10 Canada France Germany Italy Japan UK US Canada (*) (*) (*) (*) (*) (*) France (*) (*) (*) (*) (*) (*) Germany (*) (*) (*) (*) (*) (*) Italy (*) (*) (*) (*) (*) (*) Japan (*) (*) (*) (*) (*) (*) UK (*) (*) (*) (*) (*) (*) US (*) (*) (*) (*) (*) (*) 0 C(1) 15 Canada France Germany Italy Japan UK US Canada (*) (*) (*) (*) (*) (*) France (*) (*) (*) (*) (*) (*) Germany (*) (*) (*) (*) (*) (*) Italy (*) (*) (*) (*) (*) (*) Japan (*) (*) (*) (*) (*) (*) UK (*) (*) (*) (*) (*) (*) US (*) (*) (*) (*) (*) (*) 0 A-9/A-10

80 K-S Tests 2 C(1) 20 Canada France Germany Italy Japan UK US Canada (*) (*) (*) (*) (*) (*) France (*) (*) (*) (*) (*) (*) Germany (*) (*) (*) (*) (*) (*) Italy (*) (*) (*) (*) (*) (*) Japan (*) (*) (*) (*) (*) (*) UK (*) (*) (*) (*) (*) (*) US (*) (*) (*) (*) (*) (*) 0 C(1) 25 Canada France Germany Italy Japan UK US Canada (*) (*) (*) (*) (*) (*) France (*) (*) (*) (*) (*) (*) Germany (*) (*) (*) (*) (*) (*) Italy (*) (*) (*) (*) (*) (*) Japan (*) (*) (*) (*) (*) (*) UK (*) (*) (*) (*) (*) (*) US (*) (*) (*) (*) (*) (*) 0 C(1) 30 Canada France Germany Italy Japan UK US Canada (*) (*) (*) (*) (*) (*) France (*) (*) (*) (*) (*) (*) Germany (*) (*) (*) (*) (*) (*) Italy (*) (*) (*) (*) (*) (*) Japan (*) (*) (*) (*) (*) (*) UK (*) (*) (*) (*) (*) (*) US (*) (*) (*) (*) (*) (*) 0 A-10/A-10