Forecasting tools and procedures at the Banque de

Forecasting tools and procedures at the Banque de France FORECASTING MODELS AND PROCEDURES OF EU CENTRAL BANKS April 23, 2008, Sofia Macro analysis and forecast division, Banque de France

Main points of the presentation • General overview & a focus on the iteration of the macroeconomic and public finances division. • Revised version of OPTIM • Forecasting inflation tools • French retail sector specificities • Dealing with minimum wage • Temptative labor share equation as a guardrail to help the forecast

Forecasting procedures at the Bd. F: General overview & a focus on the iteration of the macroeconomic and public finances division. Delphine Irac Banque de France

Public Finances 1. Spring Exercise: • • End of March: annual public finances accounts delivery. Not all components (for instance no investment for local administrations. Public finances division: almost one week to build a consistent historical database on the previous year. + make a forecast of the public fi variables Delay the launching of the macroeconomic model based forecast 2. Winter Exercise: • • From September to December: public finances law project+amendments Difficult to adjust the annual forecasts of the public finances division with the quarterly public finances data (used in the model).

Monthly forecasting of French GDP: a revised version of the OPTIM model Banque de France Macro-analysis and forecasting division

Monthly forecasting of French GDP: a revised version of the OPTIM model 1. Description of OPTIM 2. Modelling strategy and data selection 3. Results 4. Conclusion

The spirit of OPTIM (1/2) Brief overview of the difft methods used in short run assessment p a. b. c. d. 2. a. b. c. Methods equation by equation: Factor models Bridge models Bayesian averaging Forecasts combination (Stock and Watson 2004) Model based method VAR, Bayesian VAR Neo keynesian model Accounting relationships

The spirit of OPTIM (1/2) • OPTIM = 1 b + 1 d • + GETS procedure

1. Description of OPTIM The main characteristics • • • Bridge model created by Irac and Sédillot (2002) New version by Barhoumi, Brunhes-Lesage, Darné, Ferrara, Pluyaud and Rouvreau (2007) Forecasts for French GDP and its components for the current quarter (and for the next one, in a forthcoming version) Based on monthly indicators (survey data and hard data) Use: SRA BMPE (joint with the macro model Mascotte) + internal conjonctural assessments, monthly

1. Description of OPTIM A revised version of the model • New equations • Main contribution of the revised model: Monthly forecasts (previously quarterly forecasts) • Systematic data selection using Gets

2. Modelling strategy and data selection Modelled components (1/3) • French GDP quarterly growth rate + GDP components quarterly growth rate • Some components are not modelled (production of non market services, immaterial investment, changes in inventories) • Aggregation with equations

2. Modelling strategy and data selection Modelled components (2/3) A. On the demand side: • Household consumption, computed by aggregation of the forecasts for: Household consumption in agri-food goods Household consumption in energy Household consumption in manufactured goods Household consumption in services • Government consumption • Investment, computed by aggregation of the forecasts for: Corporate investment in machinery and equipment Corporate investment in building Household investment Government investment • Exports • Imports

2. Modelling strategy and data selection Modelled components (3/3) B. On the supply side: • Total Production, computed by aggregation of the forecasts for: Production of agri-food goods Production of manufactured goods Production of energy Production in construction Production of market services C. Total GDP is forecast using a regression on total production.

2. Modelling strategy and data selection Monthly exercises • • • 3 forecasts for each quarter After the publication of Insee and EC surveys and before the ECB « monetary » Governing Council When data are missing for some months of the last quarter, the value for the quarter is computed as the 3 -month moving average of the last available observations

2. Modelling strategy and data selection The data set (1/3) • Monthly or higher frequency data • Soft (survey) data and hard data • Recent information (less than 2 months)

2. Modelling strategy and data selection The data set (2/3) Source Data type Frequenc y Publication lag Quarterly National Accounts Insee Hard Quarterly 45 Industrial Production Index Insee Hard Monthly 40 Consumption in manufactured goods Insee Hard Monthly 25 Eurostat Hard Monthly 20 New cars registrations CCFA Hard Monthly 2 Electricity consumption RTE Hard Daily 1 Declared housing starts Ministry of Equipment Hard Monthly 30 Business surveys in industry Banque de France Soft Monthly 15 Business surveys in retail trade Banque de France Soft Monthly 15 Business surveys in services Banque de France Soft Monthly 15 Business surveys in industry Insee Soft Monthly 0 Business surveys in retail trade Insee Soft Monthly 0 Business surveys in services Insee Soft Monthly 0 Business surveys in construction Insee Soft Monthly 0 Consumer surveys Insee Soft Monthly 0 Survey on public works FNTP Soft Monthly 35 Name HICP in agri-food

2. Modelling strategy and data selection The data set (3/3) Nov. IPI Dec. Bd. F survey Dec. cons. in manuf. goods Dec. IPI Jan. Bd. F survey Jan. cons. in manuf. goods Feb. Insee and EC surveys January February 1 st forecast for Q 1 Q 4 GDP release Jan. IPI Feb. Bd. F survey Feb. cons. in manuf. goods Mar. Insee and EC surveys Mar. IPI Mar. Bd. F survey Mar. cons. in manuf. goods Apr. Insee and EC surveys March April 2 nd forecast for Q 1 Apr. Bd. F survey Apr. cons. in manuf. goods May Insee and EC surveys May 3 rd forecast for Q 1 GDP release

2. Modelling strategy and data selection General specification of the equations • Autoregressive-distributed-lag (ADL) bridge equations

2. Modelling strategy and data selection Data selection procedure (1/2) • • • Systematic data selection using Gets Preselection of explanative variables strongly correlated with the modelled variable but not with each other No mix between similar data sources No use of synthetic survey indicators Selection of a first set of equations with an emphasis on economic content Final selection with rolling forecasts, taking into account the data availabilty

Monthly forecasts • Optim: v Same equation for a given quarter v RHS missing explanatory variables are estimated using ad hoc methods (average of the observed months etc. ) v Main drawback: very likely to miss turning points • Alternative methods (e. g. INSEE) v Different equations for different months v The equation specification is optimized w. r. t the set of data that are available when the fcsts is implemented v Drawback: more difficult to analyse/justify fcsts revision since change in the equation and change in the model

3. Results Root Mean Squared Errors

4. Conclusion • Satisfying results given the comparisons with benchmarks • Next step: future quarter forecasts • Problems concerning the aggregation of forecasts for GDP components

Forecasting inflation : 3 tools according to the horizon Banque de France Macro-analysis and forecasting division

3 tools according to the horizon of analysis • Very short term (3 months ahead) - NIPE – Very detailed analysis (≈ 50 components) – Unconditional projections (persistence of inflation) • Short term (1 year ahead) - NIPE – Detailed sectored analysis (≈15 components) – Conditional to import prices, wages… • Medium term (2 years ahead) - BMPE – HICP and HICP excluding energy – Value added Prices and Import deflator

Very short term (3 months ahead) • Available information • Non conditional forecast: Stochastic process Zt, observable from t = 1 until t = T Forecast : • SARMA processes (Use of tramoseats)

Very short term (3 months ahead) Food • Each main component is modelled with an equation: ü Meat product HICP ü Wheat product HICP ü Milk product HICP ü Oil product HICP ü Non-alcoholic HICP

Very short term (3 months ahead) Food • Explanatory variables: ü Wholesale prices ü Producer prices ü Lagged variable

Short term (1 year ahead) - NIPE List of components (components asked for the NIPE in blue) 0. 207 FOOD Econometric model 0. 085 Unprocessed food (meat, fish, vegetables, fruit) ARMA with fixed seasonal effects 0. 122 Processed food 0. 099 - Processed food excluding tobacco ECM 0. 023 - Tobacco Hikes according to announcement 0. 298 MANUFACTURED GOODS ECM 0. 079 ENERGY 0. 044 Oil products ECM 0. 035 Other energies (gas, electricity) ARMA 0. 416 SERVICES 0. 027 Communications Least square on seasonal dummies 0. 362 Private services ECM 0. 014 Rail and road transports Least square on seasonal dummies 0. 008 Air transports Regression on brent prices

Short term (1 year ahead)- NIPE The underlying HICP • The underlying HICP is composed by three main sectors =>Private services HICP (Housing services, Healthcare, Restaurants) =>Processed food HICP =>Industrial goods HICP • For each sectors, an Error Correction Model where yearon-year inflation is supposed to be consistent with an I(1) process. • Exogenous variables come from Mascotte and ECB assumptions • Sectors depend on different factors

Short term (1 year ahead) – NIPE Sectors dependent on different factors • Manufactured goods: • • Wages Prices of raw material Import prices Capacity utilization rate

Short term (1 year ahead) – NIPE Sectors dependent on different factors • Private services: • Wages • Unemployment rate

Short term (1 year ahead) – NIPE Sectors dependent on different factors • Processed food prices A model with two equations: - Domestic Agricultural prices depend on international food prices - Processed food prices depend on domestic agricultural prices, unit labor cost and the capacity utilization rate.

Short term (1 year ahead)- NIPE The energy HICP • The energy HICP is disaggregated into two components =>oil products =>gas and electricity • Oil products are modelled in two steps =>Oil products HICP without taxes is modelled with an ECM with the price of the brent as exogenous variable =>Taxes are included to take into account their nonlinearity • Gas and electricity prices are modelled via seasonal and non seasonal dummies

Short term (1 year ahead)- NIPE The unprocessed food HICP • Four sub-indexes =>meat products HICP =>fish products HICP =>fruit HICP =>vegetable HICP • ARMA with fixed seasonal effects

Short term (1 year ahead)- NIPE Research on a new inflation forecasting model • A need to reassess the equations: • A longer period • Changes in quarterly national accounts • Investigation on the best level of aggregation • New exogenous variables such as producer prices

Short term (1 year ahead) – NIPE Residuals – Industrial goods

Short term (1 year ahead) – NIPE Residuals – Private services

Short term (1 year ahead) – NIPE Industrial goods

Short term (1 year ahead) – NIPE Private services

The French retail sector specificities: Banque de France Macro-analysis and forecasting division 18 March 2008

Introduction • The reform of the retail sector: => From 1996, a sector with low competition => From January 2006 to January 2008, three reforms have changed the competition environment • French sellers/retailers negociation context

The reform of the retail sector 1. A sector with low competition • The legislation =>The “Raffarin law” (1996) : An authorization is needed to open a retail shop =>The “Galland law” (1996) : It is forbidden to sell beneath the unit cost • As a result, from 1996 to 2004, the inflation in processed food prices is higher in France than in the euro area

The reform of the retail sector 1. A sector with low competition The inflation in processed food prices is higher in France than in the euro area from 1996 to 2004 Processed food year on year inflation rate 1996 2004

The reform of the retail sector 1. A sector with low competition Even if the sector was competitive, retailers would have positive margins thanks to commercial services Producer Retailer Pays commercial services Receives commercial services Margin 2 Final consumer Receives Pays unit cost sell price Receives sell price Margin 1

The reform of the retail sector 2. From 2004: a new competition environment • In January 2004, sellers and retailers are urged by the French government to negociate their prices down • From January 2006, a new breakeven point is defined: commercial margins are partly deductible from the unit cost. • From January 2006 to January 2008, the amount of commercial margins that is deductible

The reform of the retail sector 2. A new competition environment • Consequently: the inflation in processed food was below 1% from 2005 to July 2007

The negociation context Prices are negociated at fixed dates • Negociation rounds occur at fixed dates. • Negociations from producers to retailers mostly occurs in January and February. • In the milk market, prices are fixed four times a year by a national syndicate of milk producers. As a result, producer prices are less volatile. • Menu-costs: The impact of the increase in

Dealing with minimum wage indexation in the forecast

Date SMIC = Worker type 1 (w 1) is paid: Worker type 2 (w 2) is paid: Worker type 3 (w 3) is paid: Latest wage agreement issued: Year N MARCH Wage agreement 1000 € 1009 € 1020 € W 1 =1000€ W 2 = 1009€ W 3 = 1020€ Year N JULY SMIC raise 1010 € 1010€ (SMIC effect indexation) 1020 € (no indexation) W 1 =1000€ W 2 = 1010€ W 3 = 1020€ Year N+1 MARCH Wage agreement 1010 € 1010€ ? See below

Wage indexation in France • In March N+1, workers of types 2 and 3 will try to catch up with the increase in the wages of workers of type 1 (+1% compared to the previous agreement). • However, there is no reason why there will be full indexation of the other wages on the raise of the SMIC. Therefore, the bargaining process could typically end up in a statu quo agreement such as: • w 1 : the SMIC (1010 €) • w 2 : 1010 or 1015€ • w 3: 1020 or 1025 €

Estimation of a labor share equation as a guard

Labor share equation • • • Benchmark equation: Y=F(K, BL)=Kf(l) With l=BL/K w/p=Bf’(l) Labor share=s=lf’(l)/f(l) • Bentolila & Saint. Paul: • s=g(k) with k=capital output ratio (SK) schedule

Shifts in the SK schedule • Changes in oil prices: shifts in the SK schedule • Effect of mark-up: S=µ-1 g(k) • Increase in workers bargaining power shift the SK schedule upwards

Estimations • Dependent variable: 1 -labor share (TM) Coeff Student DLOG(TM 1(-2)) 0. 152061 2. 487627 D(social wedge) -1. 409215 -6. 079143 DLOG(PRODT) 3. 349982 9. 439873 D(CURBDF(-1)) 0. 464890 1. 690725 D(D(Urate. O_BIT)) 0. 031071 2. 136494 LOG(TM 1(-1)) -0. 090613 -3. 838195 LOG(Globalization(-1)) 0. 125471 4. 370235 LOG(Minimum wage(-1)) -0. 040488 -4. 015430 Interest rate(-1) 1. 029257 5. 166875

Contributions