Banque nationale de Belgique SA · EURONEXT:BNB
Banque nationale de Belgique SA — NBB conference 'Understanding inflation dynamics' 2018
18 slides from Banque nationale de Belgique SA's conference presentation, Oct 26, 2018
Slide 1: Price indices with variable mark-ups and changing variety
Price indices with variable mark-ups and changing variety Thomas Demuynck1 Mathieu Parenti2 NBB conference, Brussels ’Understanding inflation dynamics’ October 26th, 2018 1 ECARES, ULB 2 ECARES, ULB and CEPR
Slide 2: Idea
Idea • Price-index variations aim at measuring changes in the standard of living • i.e. Money needed to obtain the same level of utility given new prices. • Constructing a (micro-founded) price-index requires both prices, consumed quantities. • Prices and consumed quantities are influenced by, • Cha
Slide 3: The problem with estimating price-index changes
The problem with estimating price-index changes • Implicit to the previous definition is that preferences don’t change • and thus that the demand system does not change. • Assumption easily violated by the data −→ so demand shocks are added (a.k.a. OLS residuals) ... • ... which assume that preferen
Slide 4: Redding and Weinstein (2018): reinterpreting demand shocks
Redding and Weinstein (2018): reinterpreting demand shocks • Introduce preference (demand) shifters by introducing vector ϕ ∈ Rn++ . ϕ1 q1 ϕ2 q2 u(q) → u(ϕq) where ϕq = . . .. ϕn qn • Expenditure function, e(p, u, ϕ) = min p 0 q s.t. u(ϕq) ≥ u, q 0 p = min q̃ s.t. u(q̃) ≥ u, set (q
Slide 5: Demand: introducing variable markups
Demand: introducing variable markups • Redding and Weinstein (2018) focus on CES-preferences i=N X σ−1 U= (ϕi qi ) σ i=1 • Leads to an iso-elastic demand when N is large −→ ill-equipped to capture variations in markups.
Slide 6: Demand
Demand • We assume instead a translog expenditure function as in Feenstra (2010) and Feenstra and Weinstein (2017). 1 ln(e(p/ϕ)) = α0 + α0 ln(p/ϕ) + ln(p/ϕ)0 Γ ln(p/ϕ). 2 Pn • with i=1 αi = 1, Γ is symmetric and each row/column sums to zero. • Then, si (p/ϕ) = α + Γ ln(p/ϕ).
Slide 7: Price index
Price index • Exact price index given by an augmented Törnqvist index (Diewert, 1974) n X s1 + s2 ln(e(p 1 /ϕ1 )) − ln(e(p 0 /ϕ0 )) = i i (ln(pi1 /ϕ1i ) − ln(pi0 /ϕ0i )). i=1 2 • Complication w.r.t. CES case: n is the total number of potential goods, some of which might not be available in a certai
Slide 8: Shadow prices
Shadow prices • Reservation prices are a function of the prices of all available varieties through the demand system. • Notations: I set of common goods available in both periods; sit expenditure share on i at time t; sIt total expenditure share on I at time t. • New expression of our price-index: n
Slide 9: Estimation of γ - Price equation
Estimation of γ - Price equation • Oligopolistic competition. • πi (p) = Di (p)(pi − ci ) where Di (p) is the demand for good i. • Given the translog specifications, the FOC (taking expenditures as given) are, si n pi = ci 1+ , γ(n − 1) pi − ci si n si = ≈ . ci γ n−1 γ 1 • Average mark-up is equal t
Slide 10: Estimation of γ - Identifying assumption
Estimation of γ - Identifying assumption • Idea, independence between demand and supply shocks (Feenstra 1994) (ϕ1i /ϕ0i ) ⊥ (cj1 /cj0 ) for all i, j ∈ I . • Problem: ϕ1i /ϕ0i is not identified from data (given γ). • We make a “difference in difference” (time periods + goods) • With our pricing equa
Slide 11: Estimation
Estimation • Finite sample moment condition, 1 X h i θ̂|I | = ln(ϕ1i /ϕ0i ) − ln(ϕ1j /ϕ0j ) ln(ci1 /ci0 ) − ln(cj1 /cj0 ) = 0. |I |(|I | − 1) i,j∈I • This is a U-statistic (van der Vaart, 1998), p |I | (θ̂|I | − θ) ∼a N (0, 4σ 2 ), where σ 2 is the variance of, h i E ln(ϕ1i /ϕ0i ) − ln(ϕ1j /ϕ0j ) ln
Slide 12: Validity check
Validity check Monte Carlo simulation (1000 draws). n |I | rejection rate (nominal level 5%) 30 10 0.164 90 30 0.087 150 50 0.068 300 100 0.060 900 300 0.056.
Slide 13: Application/Illustration
Application/Illustration • ACNielsen homescan panel for Denver area (Aguiar and Hurst, 2007). • Grocery packed goods purchases for a large number of households from Jan 1993 until March 1995 • Focus on 1993/1994 price indices • We aggregate over consumers and year. Total value purchased of every goo
Slide 14: Application/Illustration
Application/Illustration • Summary statistics over the 30 groups. mean std min max γ 0.083257 0.070692 0.012714 0.31657 mean mark-up 0.25797 0.098305 0.098714 0.54176 |I | 93.7 78.480 30 321
Slide 15: Price decomposition
Price decomposition • Price decomposition ln(e(p 1 /ϕ1 )) − ln(e(p 0 /ϕ0 )), n 1 X (si0 )2 − (si1 )2 1 X (si0 )2 − (si1 )2 = − , variety effect γ i=1 2 γ i∈I SI0 + SI1 X s0 + s1 + i i ln((pi1 /ci1 )/(pi0 /ci0 )), mark-up effect i∈I SI0 + SI1 X s0 + s1 + i i ln(ci1 /ci0 ), cost-shift effect i∈I SI0 +
Slide 16: Example
Example 1.04 1.02 | | 1 | weer en cceneccccneccccncccccccccrccedMccercccecce 0.98 | | 0.96 | 7 0.94 | | | FP SJ a < > < » we sg Se & RS: S > SS oS
Slide 17: Price decomposition
Price decomposition • Summary statistics (in %-points) varieties mark-up supply demand total mean -0.64418 -0.74634 0.33683 0.36598 -0.86497 std 1.0592 1.7483 6.3052 3.9010 4.5696 min -3.5287 -4.8344 -9.9060 -15.8201 -11.3404 max 1.3733 1.8282 24.7810 5.1179 7.8084 • Correlation matrix (in %-points)
Slide 18: Conclusion
Conclusion • Importance of mark-ups • Decomposition of price index into different components • U-statistic based estimator for γ (for variable number of goods). • Future research • Belgian supermarket data (?) • multi-product firms • Trade data (mulitple years/countries of origin) • Comparison with
Banque Nationale De Belgique — NBB conference 'Understanding inflation dynamics' 2018 on 6ix