Can 123 Variables Say Something About Inflation in Malaysia?

Can 123 Variables Say Something About Inflation in Malaysia? Kue-Peng Chuah 1 Zul-fadzli Abu Bakar Preliminary work - please do no quote First version: January 2015 Current version: April 2017 TIAC - BNM Workshop Monetary Policy In Theory and Practice Session 2: Measuring Inflation Dynamics 23 May 2017 1 Presentation by Kue-Peng Chuah, chuahkp@bnm.gov.my The opinions expressed in these slides and presentation are solely the responsibility of the authors and do not necessarily reflect the views of Bank Negara Malaysia. 1 / 31

Outline 1 Introduction What we do in the paper Contribution Motivation 2 Empirical estimations Data Methodology 3 Results and conclusion 2 / 31

Outline 1 Introduction What we do in the paper Contribution Motivation 2 Empirical estimations Data Methodology 3 Results and conclusion 3 / 31

What we do in the paper Create a new measure of core inflation for Malaysia Measure inflationary pressure using the underlying inflation gauge UIG Summarise a pool of information for a large heterogeneous dataset (123 variables) - beyond price data Use the generalised dynamic factor model (DFM) - a unified systematic and efficient approach 4 / 31

Contribution to the literature Numerous measures of core inflation have been proposed No single best measure of core inflation - each measure has benefits and costs (e.g. CPI ex food and energy) UIG is a new measure to gain traction in central banks Replication of Amstad et al. (2009, 2013 and 2014) Switzerland: > 400 variables USA: > 300 variables China: > 400 variables Others: Canada (Khan et al. 2013), UK (Kapetanios, 2002), euro area (Hahn, 2003 and Cristadoro et al., 2005), New Zealand (Giannone and Matheson, 2007), Iceland (Einarsson, 2015) 5 / 31

Central banks monitor a wide range of indicators Keep Calm we have a plan to fight Inflation Monitor a huge dataset to monitor and forecast inflation Use models to help us interpret large dataset 1 Data-driven analysis Prices: Demand-side: Supply-side: Overall real activity: External: Financial markets: Surveys: >500 items in the CPI basket Labour market, monetary and financial conditions Capacity utilisation, IPI, PPI GDP growth and output gap World growth and inflation, global prices Stock market, bond yields, exchange rates Consumer and business sentiments, inflation expectations, analyst forecast 2 Model-based analysis and forecasting Phillips curve <10 variables (TS) Components analysis <30 variables (TS) Principal component analysis (PCA) >50 variables (CS) Dynamic factor model (DFM) >100 variables at mixed frequency (CS+TS) 6 / 31

Measuring core inflation Eliminate idiosyncratic noise and some volatility 1 2 1 2 Cross section (CS) Eliminate SR volatility to smoothen Time series (TS) CS + TS 3 Exclude volatile items i.e. food and energy Re-weight items in CPI that are volatile or price administered Smoothing / filtering Univariate MA Trimmed mean Filters: HP, band-pass, Kalman Multivariate SVAR Kalman Principal Component Analysis (PCA) Dynamic factor model (DFM) 3 Common shortcoming : consider a limited fraction of information or a homogenous data set 7 / 31

Outline 1 Introduction What we do in the paper Contribution Motivation 2 Empirical estimations Data Methodology 3 Results and conclusion 8 / 31

Data Balanced panel of monthly data from Jan. 2000 to Dec. 2014 Prices 54 Real activity 34 Labour 1 Money and credit 16 Financial 15 International 3 x 1 x 2 x 3 x 111 t = 1 t = 2 Real activity Prices t = T Prices Labour Financial Real activity Money International 9 / 31

Data Broad and large dataset (nominal and real variables) These relevant data are monitored for surveillance Standardise each variable prior to estimation 10 / 31

Methodology DFM as developed by Forni et al. (2000 and 2005) and applied in Amstad et al. (2014) Smoothing is done to split noise from trend using Fourier Transformation (frequency domain) Essentially, the DFM summarises the entire dataset into a single indicator called the UIG The UIG is a common factor - unobservable - affecting all the variables 11 / 31

x 1 t = 1 t + ε 1 t = b 1 f 1,t + b 2 f 1,t 1 + ε 1 t common idiosyncratic x 2 t = 2 t + ε 2 t = c 1 f 1,t + c 2 f 1,t 1 + ε 1 t 12 / 31

Methodology DFM imposes assumptions such that each variable can be decomposed into two unobservable components Common component represented by sum of unobservable factors (f ) - strong correlation with whole dataset Each variable has different weights (b) - react differently to the common shock 13 / 31

Methodology Set frequency at [12 months] to capture shocks that persist (transitory vs. persistent shocks) Set number of common factors at [two] - sufficiently reflect dataset, selection criteria using Bai and Ng (2008) Set lag to [one] (distributed lag structure) With new data release, DFM allocates a new set of weights when extracting underlying common factor to estimate the UIG 14 / 31

SR inflation signal Unobserved Use all information in entire panel MR/LR UIG idiosyncratic Unobserved Noise 15 / 31

Benefits of DFM Embracing large heterogenous dataset (e.g. granular and rich, big data) Pool models vs. pool information Agnostic about structure of economy, let data speak (applicable to EME) Reliable or clearer signal to guide policy (common vs. idiosyncratic) 16 / 31

Outline 1 Introduction What we do in the paper Contribution Motivation 2 Empirical estimations Data Methodology 3 Results and conclusion 17 / 31

10 8 Y-o-Y growth, % UIG vs. Headline Inflation 2001: 2014 6 4 2 0-2 -4 Mar-01 Aug-01 Jan-02 Jun-02 Nov-02 Apr-03 Sep-03 Feb-04 Jul-04 Dec-04 May-05 Oct-05 Mar-06 Aug-06 Jan-07 Jun-07 Nov-07 Apr-08 Sep-08 Feb-09 Jul-09 Dec-09 May-10 Oct-10 Mar-11 Aug-11 Jan-12 Jun-12 Nov-12 Apr-13 Sep-13 Feb-14 Jul-14 UIG Headline 18 / 31

8 7 6 5 4 3 2 1 0-1 -2 Mar-01 Y-o-Y growth, % Aug-01 UIG vs. Core (exclusion) Inflation 2001:2014 Jan-02 Jun-02 Nov-02 Apr-03 Sep-03 Feb-04 Jul-04 Dec-04 May-05 Oct-05 Mar-06 Aug-06 Jan-07 Jun-07 Nov-07 Apr-08 Sep-08 Feb-09 Jul-09 Dec-09 May-10 Oct-10 Mar-11 Aug-11 Jan-12 Jun-12 Nov-12 Apr-13 Sep-13 Feb-14 Jul-14 UIG Core 19 / 31

8 7 6 5 4 3 2 1 0-1 -2 Mar-01 Y-o-Y growth, % Aug-01 UIG vs. PCA 2001: 2014 Jan-02 Jun-02 Nov-02 Apr-03 Sep-03 Feb-04 Jul-04 Dec-04 May-05 Oct-05 Mar-06 Aug-06 Jan-07 Jun-07 Nov-07 Apr-08 Sep-08 Feb-09 Jul-09 Dec-09 May-10 Oct-10 Mar-11 Aug-11 Jan-12 Jun-12 Nov-12 Apr-13 Sep-13 Feb-14 Jul-14 UIG PCA 20 / 31

8 7 6 5 4 3 2 1 Y-o-Y growth, % UIG vs. Min-Max of 6 Core Inflation 2001: 2014 0-1 Min-max of core-6-2 Mar-01 Aug-01 Jan-02 Jun-02 Nov-02 Apr-03 Sep-03 Feb-04 Jul-04 Dec-04 May-05 Oct-05 Mar-06 Aug-06 Jan-07 Jun-07 Nov-07 Apr-08 Sep-08 Feb-09 Jul-09 Dec-09 May-10 Oct-10 Mar-11 Aug-11 Jan-12 Jun-12 Nov-12 Apr-13 Sep-13 Feb-14 Jul-14 UIG Min Max 21 / 31

UIG (1.38) UIG -- PCA (1.31) Core (1.01) Core-6 (0.97) PCA 0.78 -- Core 0.74 0.93 -- Core-6 0.79 0.96 0.97 Headline 0.87 0.72 0.71 0.78 Headline (1.59) 10 8 Y-o-Y growth, % UIG vs. Headline vs. Core Inflation 6 4 2 0-2 -4 Mar-01 Aug-01 Jan-02 Jun-02 Nov-02 Apr-03 Sep-03 Feb-04 Jul-04 Dec-04 May-05 Oct-05 Mar-06 Aug-06 Jan-07 Jun-07 Nov-07 Apr-08 Sep-08 Feb-09 Jul-09 Dec-09 May-10 Oct-10 Mar-11 Aug-11 Jan-12 Jun-12 Nov-12 Apr-13 Sep-13 Feb-14 Jul-14 UIG Core Headline 22 / 31

Desirable properties of the UIG Smoother than headline inflation without losing too much information (e.g. global commodity price boom in 2000s) Unbiased such that it has the same mean as headline inflation Tracks inflation dynamics (turning points) 23 / 31

Why the UIG is useful Reliable Timeliness Monetary Policy Forecast inflation 24 / 31

More reliable signal of turning points Headline too volatile, core tends to ignore a lot of information Use information across variables and time (CS and TS) Reliable Timeliness Monetary Policy Forecast inflation 25 / 31

Reliable Timeliness Daily updates to monitor and forecast inflation when new data is released (nowcasting) Monetary Policy Forecast inflation 26 / 31

Reliable Timeliness MP transmission has long, variable and uncertain lags, should not react to noise or idiosyncratic effects Exclude information not related to policymakers decision-making time horizon Monetary Policy Forecast inflation 27 / 31

Reliable Timeliness Monetary Policy Forecast inflation Robust forecasting performance Mixed frequency model 28 / 31

Conclusion (last slide!) UIG contains desirable properties as a measure of core inflation - promising preliminary results UIG can be developed further as a regular feature to complement inflation toolkit of policymakers Future work on UIG Expansion of dataset (300 variables) Robustness analysis Forecasting performance (horse race) 29 / 31

Thank you Comments welcome chuahkp@bnm.gov.my 30 / 31

References Amstad, M. and Huan, Y. and Ma, G. (2014). Developing an underlying inflation gauge for China. BIS Working Paper No. 465. Bai, J. and Ng, S. (2002). Determining the number of factors in approximate factor models. Econometrica. Bank Negara Malaysia (2008). Core inflation: measurements and evaluation. Bank Negara Malaysia Annual Report, Chapter 3. Bernanke, B., and Boivin, J. (2003). Monetary policy in a data-rich environment. Journal of Monetary Economics. Chuah, K., Chong, E., and Tan, J. (2015). Global commodity prices and inflation dynamics in Malaysia. Bank Negara Malaysia Working Paper No. 5/2015. Forni, M., Hallin, M., Lippi, M. and Reichlin, L. (2000). The generalized dynamic-factor model: identification and estimation. Review of Economics and Statistics. Forni, M., Hallin, M., Lippi, M., and Reichlin, L. (2005). The generalized dynamic factor model. Journal of the American Statistical Association. Hahn, E. (2002). Core inflation in the Euro Area: an application of the generalized dynamic factor model. Center for Financial Studies Working Paper No. 2002/11. Kozicki, S. (2001). Why do central banks monitor so many inflation indicators?. Economic Review Federal Reserve Bank of Kansas City. 31 / 31