1. IV: INTEGRATION OF GOODS MARKETS LECTURE 11: EMPIRICAL TESTS OF PPP (PURCHASING POWER PARITY) Motivating questions: How integrated are goods markets internationally? How rapidly do prices adjust?

3. Prof. Jeffrey Frankel, Harvard Kennedy School

4. API120 - Prof. J.Frankel PPP: EMPIRICAL QUESTIONS Does PPP hold: What is the estimated speed of adjustment to the LR? Are PPP deviations: related to variation in nominal exchange rates? - Can one infer causality? related to geography? - To distance? To borders? Does the Law of One Price hold better in some sectors than others? In the long run? Maybe. In the short run? No.

5. API120 - Prof. J.Frankel PPP holds well in hyperinflations (a sort of long run): The cumulative change in E corresponds to the cumulative change in CPI.

6. API120 - Prof. J.Frankel Three useful kinds of evidence: (1) The pattern of movement in real exchange rates: band or threshold Random Walk trend AR. (2) Effects of exchange rate regime on variability in Q. (3) Tests of Law of One Price for narrowly defined goods. PPP fails badly in the short run. Why?  ΔE with sticky prices? So ΔE => ΔQ ?  Or does the same exogenous real ΔQ – coming from productivity or terms-of-trade shocks – show up as ΔE or ΔP, depending if on floating vs. fixed?

7. (1) Four patterns of deviation from PPP and their likely origins: a)Band <= barriers to trade b)Random walk <= shifts in terms of trade c)Trend <= Balassa- Samuelson effect d)Autoregression <= sticky prices. Band Random Walk Trend Autoregression Q ≡ API120 - Prof. J.Frankel

8. Specification of PPP test as an AutoRegressive process δ ≡ AR coefficient; (1-δ) ≡ speed of adjustment. H 0 : δ = 1 (random walk, or unit root) H 1 : δ = 0 (full adjustment to PPP) H Alt : 0 < δ <1 (gradual adjustment to PPP). Common finding in tests of 1980s: can’t reject H 0. True problem: Insufficient power in the tests, due to insufficient data. Since 1990, studies have sought more data. API120 - Prof. J.Frankel where q ≡ log of real exchange rate; u t ≡ random disturbance with E(u t ) = 0. q t = k + δ q t-1 + u t

9. API120 - Prof. J.Frankel With 100 or 200 years of data it is not hard to reject a random walk, i.e., to detect regression to the mean. qtqt

10. API120 - Prof. J.Frankel Studies with long time series: Time Period Estmt. δ Speed of Adjustment Half-life (years) JF (1990) 1869-1987.84.164 Updated WTP (2007) 1791-2005.88.12 (s.e.=.05) 4 Lothian & M. Taylor (1996) 2 centuries 3-5 Alan Taylor (2002) 19 currencies 1870-1996.79.21 (s.e.=.01) 3.4 – 4.1 Papell & Prodan (2005) 1870-1998 1-2 Cross-country Number of panel data studies: countries Frankel & Rose (1995) 150 4 Wei & Parsley (1998) 14 4-5 Choi, Mark & Sul (2004) 21 5.5

11. API120 - Prof. J.Frankel Taylor spliced together 100+ years of data for 20 currencies: 1870-1996 One lesson: reversion to LR Some have trends.

12. (2) Var(e t ) and Var (q t ) are correlated across regimes. Is it coincidence? No, it can’t be: Every time a regime switch raises variability of nominal exchange rates, it also raises variability of real exchange rates. · Pre- and post-1973 (Fig. 19.4) · Inter-war period ( Eichengreen, 1988 ): 1922-26 float vs. 1927-31 fix · Post-war regimes ( Mussa, 1986 ): Canadian float in the 1950s · A century of PPP ( A.Taylor, 2002 ) : 1870-1914 Gold standard 1914-45 Interwar 1946-71 Bretton Woods 1971-96 Float

13. When nominal exchange rate variability (¥/$) went up with floating, real exchange rate variability went up in tandem. Coincidence? Figure 19.4 Nominal & real exchange rates both became more volatile after 1973. 1973 API120 - Prof. J.Frankel

14. The final nail in the coffin: Exchange rate variability across a century of regimes Variability of real exchange rate Variability of nominal exchange rate Again, each time a more flexible regime raises nominal variability, it raises real variability too. 1870-1996 Each observation is a country-regime. (Adapted from A.Taylor, 2002)

15. API-120, Prof.J.Frankel (3) Tests of the Law of One Price (LoOP) 1.NonTraded Goods & Services, e.g., haircuts & housing have little scope for arbitrage. 2.Commodities – oil, minerals, & agriculture. -- where arbitrage can potentially work very well. 3.Disaggregated manufactures 4.Retail: In reality, even TGs have a NTG component (distribution & marketing), & vice versa. 5.Pass-through of import prices. McDonalds hamburgers. The Economist Parsley & Wei (2003). Some models make the TG/NTG line endogenous : Bergin (2003); Ghironi & Melitz (2004)

16. Professor Jeffrey Frankel, Prices of nontraded services vary widely. Notice that they are lower in poorer (low-wage) countries than in high-wage countries. 1. NTGs

17. Arbitrage equalizes prices for a homogenous metal such as gold => The Law of One Price holds. G.Alessandria & J.Kaboski, 2008, “ Why are Goods So Cheap in Some Countries?” Business Review Q2 (Fed,Res,Bank of Philadelphia), Table 2. “ Why are Goods So Cheap in Some Countries? Note: India has tariffs & quotas on gold imports. { ITF-220 Prof.J.Frankel 2. Commodities

18. Professor Jeffrey Frankel, Harvard University Even in a manu- facturing sector as disaggregated and seemingly standardized as ball bearings, the relative price in Japan varies (1)widely, and (2)in correlation with the ¥/$ exchange rate. Giovannini, JIE, 1988 3. Manufactures

19. ITF-220 Prof.J.Frankel Big Macs are partly traded (ingredients) & partly nontraded (cooking & retail). Their price varies widely across countries. Jan.22, 2014 Why was the price of Big Macs… so high in Norway? than in Japan? higher in Brazil Low in India & S.Africa? 4. Big Mac s It tends to be higher in rich countries and in countries with overvalued currencies.

20. Import prices in Turkey show much more pass- through from depreciation than in the US. With standard error bands Gita Gopinath, 2015, “The International Price System,” NBER Working Paper, No. 21646, Oct. A 10% depreciation of the Turkish Lira results in its import prices in Lira rising by 9.3% one quarter after the shock and by 10.0% eight quarters after the shock [full pass-through]... In the case of the U.S. (thick dashed line) the numbers are much lower at 3.4% and 4.4%. 5. Pass-through to import prices

21. 76 developing and other countries, 1990-2001 Source: Frankel, Parsley & Wei (2012) Exchange rate pass-through to domestic prices Pass-through coefficient ≡ % change in local price resulting from a given % change in exchange rate. Pass-through is greatest for imported goods at dock, but less for prices of the same goods at retail level. Reason: local distribution & retail costs. The pass-through to prices of local substitutes is again less; and is still less to the CPI.

22. Passthrough coefficients for less developed countries > for rich, historically. Source: Frankel, Parsley & Wei (2012 )

24. Looking ahead Barriers to International Integration -- Transportation costs -- Tariffs & non-tariff trade barriers -- Border frictions -- Currencies N on - T raded G oods -- The Balassa-Samuelson Effect