Central bank intervention within a chartist-fundamentalist exchange rate model: Evidence from the RBA case

Transcription

1 Central bank intervention within a chartist-fundamentalist exchange rate model: Evidence from the RBA case Abderrazak Ben Maatoug Bestmod, Institut Supérieur de Gestion, Tunisia Ibrahim Fatnassi Fiesta, Institut Supérieur de Gestion 41, Avenue de Liberté, Cité bouchoucha, Le Bardo 2000, Tunisia Abdelwahed Omri Fiesta, Institut Supérieur de Gestion, Tunisia February 8, 2010 Abstract In this paper we study the effectiveness of central bank intervention within a heterogeneous expectations exchange rate model. The empirical evidence is conducted by applying a Markov switching approach to daily AUD/USD exchange rate and intervention data of the Reserve Bank of Australia from 1983 to Our results are supporting both chartists and fundamentalists regimes. It is shown that the two regimes are persistent and that the fundamentalist one is riskier. Besides, the Reserve Bank of Australia is exerting a stabilizing effect by their interventions. JEL classification: F31; G15; E58 Keywords: Foreign exchange; central bank intervention; regimeswitching; chartist and fundamentalist analysis Adress: abderrazak.benmaatoug@isg.rnu.tn Adress: Ibrahim.Fatnassi@isg.rnu.tn (Le conférencier) 1

2 1 Introduction An intervention occurs when a monetary authority buys (sells) foreign exchange, this action will affect the monetary base, interest rates, market expectations and intimately the exchange rate. This type of intervention is called non-sterilized intervention. On the other hand, intervention is said sterilized if the monetary authority offsets or sterilizes the effect of the foreign exchange operation on the monetary base by selling or buying domestic bonds. This sterilization aims to keep the monetary policy unchanged. The goals of sterilized interventions are not revealed by central banks; this has been extensively explored in the empirical literature (Almekinders et al., 1996; Frenkel et al., 2004; Ito and Yabu, 2007; Jun, 2008) and in papers of surveys (Neely, 2001). The main results suggest that interventions tend to be conducted in order to reduce exchange rate misalignment or to reduce foreign exchange market volatility. The effectiveness of sterilized interventions was usually evaluated on the basis of traditional macro economic channels: the portfolio balance channel and the signaling channel. With the emergence of microstructure approach of exchange rate (Lyons, 2006) and the failure of numerous empirical studies (based on the asset market approach of exchange rate) to explain short term movements of exchange rate (Lewis, 1995 and Taylor, 1995) many other studies have used to introduce microstructure based channels and microstructure variables. These studies suggested two new channels through which sterilized intervention may be transmitted: the noise trading channel (Hung, 1997) and the coordination channel (Reitz and Taylor, 2008, Taylor, 2004, 2005). Microstructure specification like order flow, Bid-ask spread, market makers, heterogeneity of traders, was also taken into account for studying the effectiveness of intervention (Beine et al., 2009, Chari, 2007, Scalia, 2008). The noise trading channel have been pioneered by Hung (1997) and is based on the functioning and the microstructure of the foreign exchange market. This channel assumes two hypotheses: noise traders 1 must prevail the foreign exchange market and the exchange rate is determined by flow market equilibrium. Once these hypotheses are satisfied, the central bank 2

3 should intervene in highly volatile market periods and keep its interventions secret. This study contributes to the noise trading channel by allowing intervention to influence the both forecasting rules of chartists and fundamentalists, thereby altering the proportion of the two groups in the foreign exchange market. We define a new criterion for the effectiveness of interventions. A central bank intervening in the foreign exchange market is considered effective if the exchange rate is driving closer to its fundamental value. The remainder of the paper is organised as follow. Section 2 describes the literature. Econometric methodology is presented in section 3. Our main empirical results concerning intervention effectiveness are reported in Section 4 before the final section concludes 2 Literature and theoretical background In the literature of financial markets, participants are classified in two groups according to different approaches of expectations; the fundamental analysis and the chartist analysis. The classification was introduced by Frankel and Froot (1986, 1990), and has been enhanced among others by Ahrens and Reitz (2004), Reitz (2005), Westerhoff (2003), De Grauwe and Grimaldi (2005), Wieland and Westerhoff (2005). Beine et al. (2009) investigated the effect of sterilized intervention in a noise trading channel with two states Markov switching model. using biweekly data, they found that interventions increase the weight of fundamentalists in the foreign exchange market and therefore exert stabilizing influence on the exchange rate. The fundamentalists behaviour tends to stabilize the market while the presence of chartists may cause destabilization 3

4 3 The econometric model 3.1 A basic chartist-fundamentalist model for the exchange rate As discussed in the last section, we consider heterogeneous traders in our model: fundamentalists and chartists. Fundamentalists, using fundamentalist rule, base their forecasts on the fundamental exchange rate; they expect that exchange rate converge to its fundamental value. While chartists, using technical analysis rules, explore paste movements of exchange rate in the future. Fundamentalists forecasting rules and chartists forecasting rules can be expressed, respectively, as follow: r f,t = ψ(e t 1 f t 1 )+ǫ f,t (1) r c,t = C(r t i )+ǫ c,t (2) Fundamentalists consider exchange rate as a reaction function to misalignements of exchange rate level with the fundamental value. r f,t is the forecasted value of the exchange rate return r t by fundamentalists, ψ is, f t 1 is the fundamental exchange rate, ǫ f,t is the error term of fundamentalists. r c,t is the forecasted value of the exchange rate return r t by chartists, C the chartists forecasting rule, ǫ c,t is the error term of chartists. In order the characterize the exchange rate behaviour of exchange rate, we choose the Markov regime-switching model. It was initially introduced by Hamilton (1989) and developed further by, among others, Engel (1994) and Dewachter (1996). In Markov switching model, dynamics of the exchange rate is governed by an unobserved state variable or a latent variable l t (l t = c for chartist regime l t = f for fundamentalists one). The indicator regime l t is parametrized as a first order Markov process and is driven by first-order transition probabilities. These transitions probabilities, in the case of two regimes, could be expressed as: 4

5 p = prob(l t = c l t 1 = c) (3) q = prob(l t = f l t 1 = f) (4) These probabilities are constant over time. In this specification, p is the probability of remaining in the chartist regime p = 1 (1 exp(π 0 )) 1 (5) p = 1 (1 exp(κ 0 )) 1 (6) 3.2 Modelling interventions with chartist-fundamentalist In the second step, we consider time varying transition probabilities (TVTP) like Filardo (1994) and Diebold et al. (1993) and we use a logistic specification. We allow central bank intervention to influence the switching regime between chartists and fundamentalists : m p = Prob(l t = c l t 1 = c,i i,t 1 ) = 1 (1+exp(π 0 π i I i,t 1 )) 1 (7) i=1 m p = Prob(l t = f l t 1 = f,i i,t 1 ) = 1 (1+exp(π 0 κ i I i,t 1 )) 1 (8) where I i,t 1 is intervention de type i at time t 1 which will be defined later. The probability to switch from the chartist to fundamentalist regime is given by (1 p), whereas the probability to switch from fundamentalist to chartist regime is measured by (1 q) i=1 5

6 4 The empirical evidence 4.1 Data and Intervention Dynamics We use daily data of AUD/USD rate exchange and amounts of interventions conducted by the Reserve Bank of Australia from December, 1983 through April, During this period, the RBA conducted 2752 interventions in the foreign exchange market. Out of the 2752 interventions, 564 involved sales of Australian dollars (i.e. purchases of USD) and 2206 involved purchases of AUD (i.e. sales of USD). For both sales and purchases in AUD, the mean absolute value of the RBA interventions was roughly 49 millions of AUD dollars (See table 1). Probabilities and conditional probabilities are given in Table 2. the unconditional probability of RBA intervention was about 43%. However, the unconditionnal probability of an RBA interventions given that the RBA had already conducted an intervention on the previous (two previous) day was higher 70% (51%). This indicates that RBA interventions tended to occur in clusters (Figure 1). The fundamental value of the exchange rate is described by the purchasing power parity (PPP) and was constructed using monthly observed consumer price indexes which was obtained from the IFS (International financial statistics). The chartists are supposed to expect that a future exchange rate increase is predicted by the proportion of the positive difference between the 14-day moving average and the 60-day moving average Table 1: Size of Interventions for the RBA All interventions Sales of AUD Purchases of AUD Mean Absolute Value of interventions

7 USD/AUD Exchange rate Millions of AUD Figure 1: AUD/USD exchange rate and RBA interventions Table 2: Probabilities and conditional probabilities Probability of an intervention 43.25% (I t 0) (on 2752 days out of 6363) Probability of an Intervention Conditional upon 70.14% an Intervention on the previous Trading Day (I t 0 I t 1 0) (on 1930 days out of 2752) Probability of an Intervention Conditional upon an 51.60% Intervention on the tow previous previous Trading Days (I t 0 I t 1 0,I t 2 0) (on 1420 days out of 2752) Note: I t is a dummy variable that take the value 1 on interventions days and 0 otherwise To introduce interventions operations into the regime-switching model, we adopted two alternatives: the first, is a dummy variable that takes 1 if the bank intervenes and 0 otherwise, in the second alternative we choose the amount of intervention its self. Besides, for each specification we add two dummy variable relative to interventions size. I Dsint, is a dummy variable that takes 1 if the amount intervention s sales, the absolute, is upper the average of sales interventions of all the sample and 0 otherwise and I Dbint is a dummy variable that takes 1 if the amount interventions purchase is upper the average of purchases interventions of all the sample and 0 otherwise. 7

8 4.2 Estimation Once of the most important criticism to the previous work investigating the dynamics of exchange rates with chartist-fundamentalist behaviour through Markov switching model is the absence of specification tests or nonlinearity tests. In fact, testing linearity versus Markov switching model is complicated because some regularity conditions for likelihood inference are not satisfied. In null hypothesis of linearity the likelihood function is nonquadratic and flat with respect to the nuisance parameters at the optimum and the scores are identically zero. Hence, the asymptotic distribution of the relevant test statistics does not posses the standard χ 2 distribution. Hansen (1992, 1996), Garcia (1998), Carrasco M (2004) proposed another alternatives that avoid these problems; In our study we choose the Hansen (1992, 1996) test. Results test, as shown in table 3, confirm the presence of Markov switching regime is our data. Table 3: Linearity versus two-states Markov switching model Hansen test Standardized LR test LR M= M= M= M= M= The p-value is calculated according to the method described in Hansen (1992,1996), using 1000 random draws for the relevant limiting Gaussian processes and bandwidth parameter M = 0, 1, 2, 3, Chartist-fundamentalist model Table 4 reports the estimates of the basic chartist fundamentalist model. the model was estimated by maximum likelihood with Rats 7.0. Parameter estimates was obtained using BFGS algorithm ans the reported t-statistics are based on the heterosckedastic consistent standard errors (White, 1982). The transition probability are above, respectively, 0.97 and 0.95 indicating high persistence of regimes. The unconditional probability of chartist regime ( P = 0.66) is higher than the one assigned to fundamentalists regime ( Q = 0.34). This is also reflected in the expected duration of each regime. 8

9 The chartist regime is expected to last up to 48 trading days whereas the fundamentalist regime have a lower duration of 23 trading days. Ours estimated variances are significant. The second moment in the fundamentalist regime is twice times as high as the variance in the chartist regime and suggest that the fundamentalist regime is riskier ( the opposite was found by Beine et al. (2009) for the European foreign exchange market). The estimates chartist and fundamentalist parameters are significant. The ψ is negative, suggesting a mean reverting behaviour of e t toward its fundamental value f t. Table 4: Chartist and fundamentalist model estimation Coefficients t-stat. C ψ π κ σc σf p q (1 p) (1 q) Q 0.66 P 0.34 Nobs 6262 LogLik *(**,***) denotes significance at the 10% (5%, 1%) level TVTP model for intervention Table 5 contains the estimates of the four model specifications with transition probability depending on the central bank intervention as described in the subsection 3.2. In the specification 1 I i,t 1 is approximated by a dummy variable whereas in the specification 2 intervention is approximated by the amount of intervention itself. For each specification we add two dummy variable relatives to the size of intervention; purchases and sales. 9

10 1.00 Chartist Regime Fundamentalist Regime Figure 2: Smoothed probabilities of chartists and fundamentalists regime As outlined in the second section, central bank are supposed to affect the exchange rates forecasting rules for both fundamentalists and chartists. In order to test this hypothesis, we examine the likelihood ration test (LRT) and the estimates of the various π i s κ i s assigned the RBA intervention in table 5. The LRT are significant, hence the hypothesis that the central bank intervention does not not affect the exchange rate is rejected for the each model specification. In Figure 2 are shown smoothed probabilities of chartists and fundamentalist regime. We can see the that during December 1993-June 1995 when the RBA did not intervene, the proportion of fundamentalist of all traders is nearly zero. We can conclude that the presence of central bank in the foreign exchange market may induce traders to adopt fundamentalist behaviour and hence exert stabilizing effect 10

11 Table 5: Chartist and fundamentalist model estimation with Central Bank interventions Model 1a Model 2a Coefficients t-stat. Coefficients t-stat. C ψ π π κ κ σc σf N obs LogLik Model 1b Model 2b Coefficients t-stat. Coefficients t-stat. C ψ π π π π κ κ κ κ σc σf N obs LogLik Signi.tests H 0 : π 1 = π 2 = π 3 = κ 1 = κ 2 = κ 3 = 0 χ 2 = [0.0000] χ 2 = 58.84[0.0000] *(**,***) denotes significance at the 10% (5%, 1%) level. Numbers in brackets are p-values for testing the joint significance of the behavior parameters. 5 Conclusion In this paper we proposed a generalisation of the noise trading channel (Hung, 1997) in order to study the effectiveness and the impact of sterilized inter- 11

12 vention on the exchange rate. Using Markov switching regime model with time varying transition probability we found that both regimes are persistent. However it is found that the the fundamentalist regime is riskier. Besides, during a period, nearly two years, when the RBA was not active in the foreign exchange market (.i.e no interventions) the fundamentalist traders were completely absent. References Ahrens, R. and S. Reitz (2004, January). Heterogeneous expectations in the foreign exchange market. Journal of Evolutionary Economics 15(1), Beine, M., P. D. Grauwe, and M. Grimaldi (2009, July). The impact of fx central bank intervention in a noise trading framework. Journal of Banking & Finance 33(7), Carrasco M, Liang H, P. W. (2004). Optimal test for markov switching. Working paper, University of Rochester. Chari, A. (2007, 08). Heterogeneous market-making in foreign exchange markets: Evidence from individual bank responses to central bank interventions. Journal of Money, Credit and Banking 39(5), Dewachter, H. (1996). Charts as signals in markov switching world. Applied Economics Letters 3(6), Diebold, F. X., J.-H. Lee, and G. C. Weinbach (1993). Regime switching with time-varying transition probabilities. Working papers, Federal Reserve Bank of Philadelphia. Engel, C. (1994). Can the markov switching model forecast exchange rates? Journal of International Economics 36(1-2), Filardo, A. (1994). Business cycle phases and their transition dynamics. ournal of Business and Economic Statistics 12,

13 Frankel, J. A. and K. A. Froot (1986). Understanding the u.s. dollar in the eighties: The expectations of chartists and fundamentalists. The Economic Record 0(0), Frankel, J. A. and K. A. Froot (1990). Chartists, fundamentalists, and trading in the foreign exchange market. American Economic Review 80(2), Garcia, R. (1998). Asymptotic null distribution of the likelihood ratio test in markov switching models. International Economic Review 39, Hamilton, J. D. (1989). A new approach to the economic analysis of nonstationary time series and the business cycle. Econometrica 57(2), Hansen, B. E. (1992). The likelihood ratio test under non-standard conditions: testing the markov-switching model of gnp. Journal of Applied Econometrics 7, Hansen, B. E. (1996). Inference when a nuisance parameter is not identified under the null hypothesis. Econometrica 64, Hung, J. H. (1997, September). Intervention strategies and exchange rate volatility: a noise trading perspective. Journal of International Money and Finance 16(5), Lyons, R. K. (2006). The Microstructure Approach to Exchange Rates. MIT Press Books. The MIT Press. Neely, C. J. (2001). The practice of central bank intervention: looking under the hood. The Regional Economist (May), Reitz, S. (2005, January). Central bank intervention and heterogeneous exchange rate expectations: Evidence from the daily dem/us-dollar exchange rate. Open Economies Review 16(1), Reitz, S. and M. P. Taylor (2008, January). The coordination channel of foreign exchange intervention: A nonlinear microstructural analysis. European Economic Review 52(1),

14 Scalia, A. (2008). Is foreign exchange intervention effective? some microanalytical evidence from the czech republic. Journal of International Money and Finance 27(4), Taylor, M. P. (2004, 02). Is official exchange rate intervention effective? Economica 71, Taylor, M. P. (2005, 02). Official foreign exchange intervention as a coordinating signal in the dollar-yen market. Pacific Economic Review 10(1), Westerhoff, F. H. (2003, December). Central bank intervention and feedback traders. Journal of International Financial Markets, Institutions and Money 13(5), Wieland, C. and F. H. Westerhoff (2005, September). Exchange rate dynamics, central bank interventions and chaos control methods. Journal of Economic Behavior & Organization 58(1),