this paper analyzes the panel data of bi-weekly surveys, conducted by the japan center for international finance, on the yen dollar exchange rate expectations of forty—four institutions for two years. there are three major findings in this paper. first, market participants are found to be heterogeneous. there are significant individual effects in their expectation institutions are found to violate the formation.
NBER WORKING PAPER SERIES FOREIGN EXCHANGE RATE EXPECTATIONS: MICRO SURVEY DATA Takatoshi Ito Working Paper No 2679 NATIONAL BUREAU OF ECONOMIC RESEARCH 1050 MassachusettsAvenue Cambridge MA 02138 August 1988 Discussions with Jeffrey A Frankel, Kenneth A Froot, Hidehiko Ichimura, Maurice Obstfeld, and Chriatoper Sims were very useful I have benefitted from comments received form participants of seminars at Harvard University, University of Minnesota, University of California at San Diego and National Bureau of Ecunuaic Research I thank the Japan Center for International Finance, especially President Iomomitsu Ohs and Vice President Shoji Ochi, Financial for having made the survey data in the Tokyo market available SES-8808828 and from support from National Science Foundation grant number is Japan Economic Research Foundation, [Nihon Keizai Kenkyu Shorei Zaidan] research program in research is of NBER's This part gratefully acknowledged International Studies Any opinions expressed are those of the author not those of the National Bureau of Economic Research NBER Working Paper #2679 August 11988 FOREIGN EXCHANGE RATE EXPECTATIONS: MICRO SURVEY DATA ABSTRACT This conducted paper analyzes the panel data of bi—weekly surveys, by the Japan Center for International Finance, on the of forty—four institutions yen/dollar exchange rate expectations for two years this There are three major findings in paper First, market participants are found to be heterogeneous are significant formation "individual Second, effects" in expectation many institutions are found to rational expectation hypothesis Third, horizons showed horizons their less violate forecasts with yen appreciation than There those Cross-equation constraints implied by the with the long short consistency the data of the forecast term structure are strongly rejected in Takatoshi Ito Institute of Economic Research 1-Jitotsubashi University Kunitachi, Tokyo, 186 Japan Introduction rational expectations have become a popular benchmark in As financial and interested in Although macroeconomic hypotheses, survey data on many domestic variables, market including participants interest rates many inflation rates, have been frequently analyzed by and become more many economists have directly measuring expectations of thinkctng investigators for example Miahkin (1983; ch 4)) it is only recently that (see, data on foreign exchange rates have become available survey been analyzed and Dominguez (1986) and Frankel and Froot (1987a.b) have exploited the survey data made available by the Money Market Service (MMS) the Amex Bank Review and the Economist Financial Report.1 The Froot surveys Frankel and Heterogeneity among the and by that were investigated by Dominguez have only their median responses reported market participants, if it existed, is aggregated out If market the with consists of homogeneous agents which share the same forecasting model common belief (priors) and information, then the median sufficiently describe the market in terms of forecasts response would if However, participants differ in their forecasting characteristics, market focusing on the median misses the most interesting questions such as whehter the differences persist or are temporary, whether the difference is correlated with the participant's traits, and whether a rationality hypothesis is more Likely to be rejected in individual data Only individual responses of survey data could answer these questions In this paper, I will use the survey data collected by the Japan Center for International Finance (JCIF) in Tokyo, which allows me the individual responses in the survey.2 In particular the JCIF data has two distinct advantages over other data used by -1- to investigate Dominguem, and set by Frankel with and Froot the JCIF data consist of First, Second not only financial no missing observations other are companies individual responses polled in the JCIF survey institutions but Therefore, chance to associate possible heterogeneity to forecasters' There are are participants "individual three major found effects" in to in findings be this are There significant Second, many institutions ace found to violate the rational expectation hypothesis them underestimated the degree of yen appreciation of with long horizons showed horizons Put differently, Third, Most forecasts less yen appreciation than ones with short market participants appear to have a "bandwagon" expectation in the short-run, "twist" a market First, expectation formation their is industry traits paper heterogeneous there but a "stabilizing" one in the long-run in forecast term structure could be "internally consistent" sense of Froot and Ito (1988)), if an iterated substitution of a forecast yields a long-term forecast, However, The (in the short-term cross-equation constraints implied by the internal consistency are strongly rejected in the data, Data Summary 2.A The Data Description The JCIF has conducted and - a telephone survey twice a month, in the the end of the month, on Wednesdays, yen/dollar exchange obtained from brokers, rate for and foreign exchange experts in 44 securities companies, Forecasts of since May 1985 one-, three- trading middle the six-month horizons are companies; companies 15 banks and export-oriented companies S life insurance companies, and import-oriented industries,4 The survey It is meticulously arranged so that all 44 companies on the permanent list respond every week, -2- ihen the mean the rndivi- following, mean across forty-four individuals at a time wrll be referred to as an the mean across individuals at a time in an (cross-section total) average: The mean across time industry group will be referred to as a group average of an individual, as across :n and the mean across time should not be confused duals the a data set is analyzed as a panel data, the (time) "average" will be referred of a group, or of the total mean of the individual, of the group or of the to avetage, respectively The JCIF calculates minimum the total average, of forty-four responses overall including average and group ICIF informs ita statistics The those who are polled, of the summary (cross-section) maximum, and also industry group averages and On the day after che survey, the standard deviations members, standard deviation, is also released to the press and other media I the will use, in addition to the panel data of forty-four companies, information part of the survey: the cross-section average public different industries: forty-four companies and group averages for securities companies (SEC) (BAN), (AyE) trading companies companies in the export industries (EXP), insurance companies in the import industries (IMP) [Sogo Shosha] dollar, so that negative movement indicates exchange rate, s(t), is measured at the closing quote banks (TRA), companies (INS), The unit is yen per yen "appreciation." in Tokyo on of and one U.S The spot Wednesday of the survey week 2.B Overview Let us illustrate the average (aggregate) expectation and the of actual exchange rates during the one-year turbulent Oroup of Five, Plaza Agreement of September 1985 -3- movement period following the In Figure 1, the solid line is the actual daily spot rate, while the base of each sequence of three arrows shows the timing of the poll and the spot exchange rate of that day Three arrows, of average Therefore, respectively, point to the forecasted exchange rate the forty four) with the one- thpee- and mistake of FIGURE several by this figure, the ABOUT HERE two preliminary observations emerge (See Ito (1987) for "news" analyeia of market, rate the expectation First, each waves of aharp appreciation does not seem to have been these of anticipated appreciation Second, the long-run expectation is not the simple extrapolation of waves,) the six-month horizons the vertical deviation of the tip of an arrow from the spot (solid) line represents an ex From (the short-run In fact, there seems to expectations be a in "twist" expectation term structure, order to quantify these observations and to extend observations to In group aggregates and provided changes Table reaponaea, Table shows that the time means of and Figure (unconditional) are expected (in percent) from the spot rate at the, time of survey for the cross- section total table) individual for average, group averages, and (in each individual a separate distribution For the purpose of discussion, the mean of forward premium (FOR) and the actual (ex Q9,f) changes of the spot exchange rate (ACT) for each horizon are reported in the same table For each horizon and each individual or group, subtracting the forward premium from the forecasts yields the implied risk premium and subtracting the changes produces the forecast errors, TABLE and FIGURE -4- about here actual the In one-month horizon, the (total) average expected a 1.4 percent yen appreciation percent to percent appreciation on a week typical The group averages ranged from 0.8 In relative to total average, the export industry was most biased toward yen depreciation and trading companies import industries were most biased toward yen appreciation individual data, one extreme predicted while the other extreme predicted 1.4 percent into Looking depreciation of 3.1 percent of appreciation The bution of individual forecasts has a nice unimodal distribution change is at about one-fifth in the distribution, and yen distri- The actual that is about 20 of participants overpredicted rhe size of yen appreciation in the percent one-month horizon The mean of forward premium (FOR) was 0.15 percent that indicating there was a risk premium for most of the participants in favor of the dollar The mean of actual changes for a one-month period was asset during the companies two years of the sample period The forecasts 2.1 percent by trading with 2.0 percent and import companies with 1.9 percent came to the actual movement close All groups and a majority of individuals did better in forecasting than the forward rate in the one-month horizon The expected appreciation of the yen in the three-month horizon for the average was that about the same as the one-month 1.4 percent, no adjustment is made with respect to the length of horizon (Note horizon.) Hence, the aggregate forecasts, one-month and three-month combined, imply that they predict little movement after the first month to the third month in the forecast horizon However, this statement will be qualified shortly when we examine the individual responses ciation in three months The actual change was percent appre- The total and group averages underpredicted the size of yen appreciation by to percentage points -5- In fact, the most yen appreciation predicted by an individual is 4.3 percent and no individual overpredicted the size of yen appreciation in the three-month horizon As in the one-month horizon, bias the (from total appreciation bias average), the export industry has a yen depreciation and the trading in the three-month horizon have a companies The implied risk that the export industry would have a negative risk premium shows dollar asset, while others have a positive risk premium as in the premium for the one-month This makes sense considering that exporters are long horizon yen in dollar assets in the medium run Wide disagreement among individuals starts to show in the It becomes a bi-mod.al distribution: forecasts yen three-month one group believes that the depreciates from the one-month to three-month in the forecast horizon while the other believes that the yen continues to appreciate bution also has long tails and The distri- Therefore, although the group averages for one- three-monthhorizons predict a yen appreciation of about the same size (1.4 percent) distributions of individual responses are quite different The last observation illustrates how important it is to have a data set with individual responses rather than one with the summary statistics For the six-month horizon, the total average practically shows that the market expects forecast the yen to return to the prevailing level at time of This is a sharp turnaround from the forecast of 1.4 percent appreciation in three months that the In fact, each of the group averages indicates the group anticipates less yen appreciation in the than either of yen the one- or three-month horizons six-month horizon The agreement among different groups in forecasting the sharp yen appreciation from the third to the sixth month is quite striking, since they differ -6- in forecasting the direction of the yen from the firsr to the third month In premium the six-month horizon, all of the groups for the dollar denominated asset have a negative risk This contrasts to the positive risk premium in the one-month horizon The distribution of individuals is almost uniform ranging from a percent depreciation to a one percent appreciation of the individual forecasts three highlighting a diversity It is clear from the figure that the degree of diversity increases as the forecast horizon is longer Again looking at the average would not give this observation Findings of this section can be aummarized and related to the of the rest of this paper First, the findings are highly heterogeneous market participants the contents suggestive of A rigorous analysis and interpretation of heterogeneity will be provided in Section Second, large forecast errors were recorded during the intermittent waves of yen appreciation after September 1985 while, the appreciation whether However, when the exchange rate was relatively stable for a one-month market Therefore, it expectation overpredicted the is not immediately clear from expectations were unbiased individual) time means without test statistics the (ACT) shows These observations are the of figures In Table 1, a comparison of group with the actual change forecast errors for each horizon amount (or average only suggestive Econometric tests on varioua forms of the rational expectation hypothesis will be conducted in Section Third, the short-horizon expectations seem to predict yen appreciation, while the long-horizon expectations seem to predict a reverse in direction Thus, their the total average and most of the group averages have a forecasts, Section investigates expectations are internally consistent whether such "twist' in twists in Wishful Expectations and Heterogeneity 3.A Econometric that Recall our special case of micro survey data and fifty-one observations individuals formation at individual j, j individual a j: u(t) is the expected exchange given rate for (3.1) a pure spot exchange rate at time t random disturbance representing The cross-section average of for individual f(I(t)) + cAVE + uAVE(t) (32) (E s(t))/J, cAVE — (E ej)/.J and uAVE contains a possible a For the 5XVE(t) is defined as 5XVE(t) f(I(t)) on based J (where in this paper J=44) is measurement error 5AVE(t) where that an individual forecast f(I(tY and the individual effect ej where si(t) is a k-step ahead forecast of the forecasts, forty-four f(I(t)) + ej ÷ uj(t) si(t) example, consists of set Suppose forecast horizon, k (suppressed notation), by data time t consists of the common structural part information, public a Issue (E Assume u1)/J constant term so that normalization, CAVE — is Then subtracting each side of (3.2) from the corresponding side of (3.1), we obtain s(t) The - 5XVE(t) — ej + (uj(t) - uAvE(t)) (3.3) estimator of individual effect ej can be obtained by regressing the lefthand side of (3.3) on a constant over the sample period (across This procedure is simple and robust It is unnecessary for econometrician to know the exact structure f(I(t)) as long as it is -8- time) the common I/IS 11/13 v,: ) 21 IZ 34) 18 ( 12/ 26 AFTER 14 10/ II St 23 YEN/I I 91 19 2(K) 214) 220 234) YciiIS Figure xPCTTD :o 2— C H G F 0- — — I B RA INS VE 0I Z0 — SEC —— EXPDR — —- INPORT — -— uncond1t.cnal 1: TABLE Epecec rarqe Tim mean of (s(t,k( — Mean of the (unconditional) e,ected change Not annualized or adjusted for May 1985 — June 1987, number of observatiOn = 51 MONTH MONTH Horizon AVE —1.420 —1.431 —0.04H BAN —1.404 —1.58 -0.957 SEC —1.097 —0.834 +Q.a21 TPA —1.956 —2.453 -0.948 EXP —0.775 —0.137 +j73 INS —1.746 —2.309 ÷0.302 IMP —1.Q37 —1.536 —0.430 FOR -0.150 -0.430 —0.957 ACT —2.064 -3.970 —11.987 Definit ion: AyE: rIONTH in place of FOR: f(t,k) in place of ACT: s(t+k) in place of in 4) Table (continued) f Distribution among individual respondents the time cnean of forecasted changes in the exchange rate aver the specified horizon HOR I ZON 1 respondent MONTH respondent P MONTH rpondent MONTH +5.0 +4.5 +4.0 U +3.5 +3.0 +2.5 — — I — — — I U U -'-2.0 +15 +1.0 +0.5 I I U U 0.0 -0.5 -1.0 — — -2.0 I -2.5 -3.0 -3.! I -4.0 U -4.! -5.0 Max Mjr, 1.41 —3.10 I I I Max 3.25 Mm —4.76 Mm 4.62 —5.20 Table 2: (ishfu1 (a) Group Deviations from the Total Average, for each norizon, — SAVE(t> = s(t.) May 1985 — HORIZON a t—tat SEC t—stat IRA t—.t.t EXP t—stat e(t) deviation Unit BAN Expectations month onth nonth DW from the cross—section average, sOBS 51 June 1987, or a RHO OW or RHO 0.530 a DW or RHO 0.371 0.017 0.28' (0.25) (2.04) —0.228 (—1.28) (4.29> —0.941 (—5.74)* (2.81) 0.305 (1.25) 0.438 (3.38> 0.561 (1.62> 0.421 >3.14> 0.743 (1.47) 0.446 (3.49) —0.536 (—4.98) DW.2.13 -1.022 (—7.56> DW=1.61 0.645 DW2.07 1.294 DW—1.62 (12.68) (8.55> INS t—stat —0.326 (—1.54) 0.474 (3.72) —0.815 (—1.93> 0.645 (5.86) IMP t—stat —0.517 (—3.76> DW—1.'.7 —0.079 (—0.29) 0.301 (2.17) • show the group at heterogeneous" Table a 1.ve1 28: Wimhful Expectations S(t,k) HORIZON the — or 0.467 (3.61> 0.435 1.832 (6.11> >3.41) 0.301 (0.54) 0.661 >5.99> —0.434 (—1.39> 0.422 (3.27) of ) II S(t,k> month OW —0.908 (—2.57)* RHO EXP—IMP t—stat 1.162 (6.73) 044—1.67 EXP—TRA t—stat 1.181 (7.98) DW—2.21 a 1.399 (6.09) 2.365 (10.31) month DU or Sonth RHO DW.1.61 0.272 (2.01) a 2.271 OW or RHO (4.19> 0.460 (3.68> 2.744 (5.09)* 0.502 (4.04) Tabl• (continued) Table 2C: Distribution of individual effects — u significant individual effects; insignificant individual effects HORIZON MONTH MONTH +5.0 I I I I +4.0 +3.5 MONTH I +3.0 +2.5 +2.0 +1.5 i +1.0 I +0.5 00 S I w r SI F 1tS! — L -2.5 I ttt I I -3.5 I U -4.0 I -3.0 —4.5 -5 • -5.5 Table 3: Idiosyncratic effects Extrapolative form ' HO: No idiosyncratic coefficient effects, b = (a1la+ing for individual effect of a constant bias) HI: No idiosyncratic effect, a—bO Lag length HORIZON coefficient or individual lag (b2 0) lag month month month month HO Hi HO Hi HO Hi HO Hi F—stat 122 103 month Hi HO (constant) month HO Hi 729 903 2.5i 2.48 120 095 2.34 18.19 133 000* 433 338 651 798 2.60 3.13 086 035 732 17.3 487 000* SEC F—stat 815 i.37 371 265 signif 037 1.41 847 253 000 1.03 984 367 699 i.Si 502 224 281 1.32 756 290 032 740 968 533 BAN signif YRA F—stat signif EXP F—stat signif 461 21.0 390 24.6 652 5.16 500 000* 535 000* 423 009 1.69 16.76 1.91 18.0 583 4.01 196 000* 161 000 .562 013 4.28 2.29 52.1 2.88 64.8 186 18.6 113 000* 067 000 .831 000* 044+ 40.5 2.16 66.44+ 557 iB.33 000* 148 000 .459 000* INS F—stat 429 2.12 signif 516 132 2.29 3.56 317 037 3.49 1.89 068 162 347 1.69 708 186 1.i2 2.71 335 056 1.46 1.14 242 345 IMP F—stat 3.68 7.70 1.36 726 signif 061 001* 249 489 1.29 2.31 262 1i0 1.73 5.4i 188 003 1.08 763 347 521 1.07 1.93 352 139 Table : Unbias.dness s(tsk) — H: a = s(t) 0, b a + b(e't,k>—s(t)) Generalized M.thod of Mo.nts, *Obe Estisates of a 51 a and b and their (standard errors) (k) HORIZON MONTH b CI1ISQ MONTH a (0.034) —0.021 —0.059 (0.015) (0.789) 2.00 —0.044 36B >0.03'.) —0.015 CHISQ 07' —0.119 0.908 10.09 (0.041) (0.741> 006 0.945 (1.226) 3.69 158 -0.118 0.220 7.90 (0.045) (0.857) 019 1.97 373 -0.054 0.706 (0.025) (0.292) 9.37 009 0.593 9.21 —0.124 (0.041> (0.327) 009 TRA —0.036 -0.793 23.35 (0.010) (0.371) 000 —0.050 0.399 (0.039) (0.947) 2.39 303 -0.110 1.038 13.3' (0.042> (0.206) 001 EXP -0.021 >0.010) 0.001 3.98 (0.747) 137 -0.058 0.896 (0.024) (0.764) 8.86 012 —0.143 1.307 12.86 (0.040) (0.443> 002 —0.030 —0.509 10.33 (0.011) >0.472) 006 —0.066 —0.264 (0.034> (0.967) 3.92 141 —0.122 0.595 9.64 >0.039> (0.665) 008 -0.023 —0.044 1.019 28.41 >0.027) (0.352) 000 —0.120 -0.016 8.é9 (0.041> (0.579> 013 —0.104 —10.360 (0.040) (6.702> —0.207 —9.159 12.25 >0.069) (3.699> 002 SEC —0.043 274 b 5.21 SAN 2.59 MONTH CHISQ a 1.167 (1.167> AVE -0.028 —0.485 (0.017) (0.969) b (0.017) INS IMP FOR 0.517 (0.502) —0.139 (0.016> (0.564> ' 76 —0.034 8.93 (0.013) —9.870 (7.432) 092 012 8.97 011 Tae 5: 0rhoqonality Past Forecast Errors as an information 5.A sC )t,k> H: a = 0, b 0, — s(tk) tested by a 4- set b)s C )t—k,k)—s(t Generalized Method of Moments, $Obm = CF = Estimates of a and b and their (standard errors) Chi—square (df =2) and its )siqn:ficance level) HORIZON (k) MONTH D CHISQ a (st.er) )st.er) Signif I MONTH MONTH a a b CHISO )st.er) (st.er) Signif )st.er) )st.er) CHISQ Signif AVE 0.005 0.049 0.881 (0.009) (0.168)0.644 0.019 0.341 6.174 (0.020) (0.191) 0.045 0.042 0.307 228.70 (0.012) (0.193) 0.000* BAN 0.006 -0.002 0.608 (0.009) (0.177) 0.738 0.016 0.359 6.467 (0.019) (0.180) 0.039 0.033 0.317 (0.010) (0.182) SEC 0.010 -0.062 1.331 (0.009) (0.131)0.51' 0.021 0.375 10.300 (0.019) (0.149) 0.006* (0.003) IRA —0.000 0.100 0.589 (0.009) (0.160) 0.745 0.011 0.337 7.'.09 (0.019) (0.140) 0.025 0.030 0.346 206.15 0.000* (0.009) (0.184) EXP 0.011 0093 3.729 (0.009) (0.171) 0.155 0.029 0.321 8.075 (0.022) (0.207) 0.018 (0.022) INS IMP FOR 0.04' 0.059 61.553 0.000 0.336 373.57 0.164) 0.000* 0.293 77.'.82 0.191) 0.000* 0.001 0.159 1.649 0.179 (0.157) 0.438 0.020 0.200 2.117 (0.021) (0.213) 0.347 0.065 (0.009) (0.023) (0.220) 0.000 0.046 (0.008) (0.184) 0.080 0.961 0.352 6.006 0.017 (0.020) (0.218) 0.050 0.047 0.244 (0.012) (0.206) 0.018 0.022 7.572 (0.010) (0.176) 0.023 0.040 0.1'.' 5.579 (0.018) (0.166) 0.061 0.048 0.208 478.02 (0.013) (0.223) 0.000* 47.248 0.000* 114.11 0.000 '.7 Forward Premium as an information 5.B: — -4 )st.er) SEC TRA EXP INS IMP FOR a + b(f(t,k)—s(t) ) + e)t 0, a BAN = b 0, tested by Generalized Method of Moments, Estimates of a and b and their standard errors) Chi—square )df 2) and its )5iqnifcance level) a — AVE s(t+k) set MONTH b (st.er) CHISQ Signif 0.0213 10.30 2.132 0.015) (7.493) 0.344 *Obs = 51 (k) HOPIZON MONTH CHISQ a b (st.er) )st.er) Signif MONTH a b CHISO )st.er) (st.rl Sqnif 0.098 (0.040) 0.221 0.069) 12.23 (6.745) 6.662 0.036 l0.1 (3.823) 15.744 0.000* 2.461 0.264 0.099 13.05 4.854 (0.040) (4.924) 0.032 (0.074) 10.85 11.342 (4.349) 0.303* 0.023 9.31 2.515 (0.015) (7.764) 0.284 0.102 11.72 7.413 (0.039) (5.758) 0.025 0.204 )0.080) (4.557) 0.000* 0.017 11.42 2.014 (0.014) (8.144) 0.365 0.087 (0.040) 0.191 8.44 13.279 (0.067) (3.565) 3.001* 0.027 9.91 4.142 >0.014) (7.004) 0.126 0.110 12.05 9.38 >0.041) (6.748) 0.009* 0.022 10.82 (0.014) (7.293) 0.018 10.43 1.749 12.07 (7.011) 0.086 11.59 0.093 11.35 5.06 0.080 4.411 (0.017) (7.922) 0.417 (0.043) (7.323) 0.110 0.0134 8.78 1.442 (0.017) (9.088) 0.486 (0.042) (6.750) 0.054 0.017 0.187 18.94 (0.009) >0.206) 0.000 >0.040) 0.104 5.843 11.36 8.97 (6.702> 0.011 0.214 8.18 45.092 0.245 11.25 23.922 (0.057) (3.022) 0.000* 0.245 12.77 17.387 (0.059) (3.154) 0.000* 0.222 11.10 13.559 (0.075) (4.204) 0.001* 0.207 (0.067) 10.14 11.485 (3.760) 3303* S.C Past echinge — H: a = rate changes: s(t+k( = a + One lag ersxon b(s)t—l '—5(t) + e(t) 0, b = 0, tested by Generalized Method of Moments, Estimates of a and b and their (standard e-rors Chi—square (df 2( and its (significance ieve) HORIZON (> MONTH a b 0H152 (st.er( (st.er) Signif MONTH a b (st.er) (st.er) HISQ Signif a o MONTH NObs = 51 CHISO (st.er) (st.er( Signif 0.042 0.306 9.504 (0.025) (0.225) 0.0094 (0.036) BAN 0.004 0.167 3.725 (0.009) (0.200) 0.155 0.041 0.233 6.440 (0.024) (0.244> 0.040 0.105 0.178 16.195 (0.038) (0.3B5) 0.000* SEC o.ooe 0.069 7.534 >0.009) (0.218) 0.023 0.047 0.389 11.327 (0.026) (0.329) 0.003* 0.120 0.318 10.274 (0.039) (0.305) 0.006* IRA —0.002 0.172 1.214 (0.010) (0.216) 0.545 0.032 0.280 5.604 (0.025) (0.264) 0.060 0.106 0.137 11.860 (0.036) (0.325) 0.003* EXP 0.010 0.232 9.290 (0.010) (0.212) 0.009* 0.054 0.382 17.283 (0.024> (0.182) 0.000* 0.132 0.230 32.705 (0.033) (0.381) 0.000* INS —0.000 0.225 4.131 >0.196> 0.127 0.033 0.315 6.940 (0.024) >0.188> 0.031 0.117 0.263 20.602 (0.035) (0.358) 0.000* 0.053 0.018 (0.220) 0.947 0.041 0.344 11.643 (0.023) (0.253) 0.003 (0.010) IMP -0.000 (0.010) FOR 0.017 0.187 18.9' >0.009> (0.206> 0.000 0.054 0.124 O.024( (0.221) 7.74 0.021 0.114 0.227 12.908 (0.358) 0.000* 0.004 0.166 3.883 (0.010) (0.203) 0.144 AVE 0.108 0.037) 13.874 0.365 (0.388) 0.001 0.108 -0.142 (0.036) (0.375) 25.93 0.000 Number of Cases in Micro Data Fail (at to reject H month month 37 26 11 18 33 1%) Reject H Cat month 1%) Table 5: (continued) 5.0 Past Exchange rate changes: Two lag version st)tk) —s)t+k) I-I: a = 0, AVE k) SEC IRA E0P INS a b st,er) )st.er 007 247 FOR c HISQ Signif star) st.er i c a st.er 38.29 043 330 -.095 (.207) 000* (.025) (.220) (.093) n355 81.10 042 270 —.152 000' 024) 235) 179) 382 028 007 256 178) 011 155 -.339 22.18 011) 205) (.187) 000' 000 227 214) 0*' (.027) (.308) (.191) -.219 11.80 032 280 001 012) (.194) (.221) 009* (.025) 264) 079) 013 —.318 31.89 011) (.197) (.2091 000+ 003 —.327 27.96 '.213) 000+ —.342 15.87 (.205) 001' 312 307 003 139 (.012) (.208) 019 259 (.011) (.182) Number 055 387 022 025) (.175) (.050) 036 376 —.243 (.024) (.195) (.084) 042 ).024 149.22 057 188 (.209) 000' (.025) (.223) among Fail to reject H (1%) H : -.121 -.255 (.074) 135 a MONTH stan Siqnif 112 103 9q) 174 002' 03*) 1.49 103 •12q 127 035) 311 ,376 118 263 037) 273) 104 P 034 268 01; 12.10 001* o.; ,0°0 18.60 000* 66.07 000* 34.94 000' 54.57 000* 131 '031) 117 19* 1.3*2) 221 189 339) 13* 3*2) (.291) 249 05' (.289) ,02 106 30* 245 034i 310 '4)3' -.088 034) (.317) 0004 11.29 001* 10S 013 032) 110 21.15 -.216 (.322) 12.3' 00*4 43.57 000* 37.05 300* i*.43 )01+ 31.2 000' Micro Data Reject 374 (.239) (.128) -.287 of cases 10138 Siqnif -.323 (.011) 012) (.133) IMP MONTH MONTH (.011) (.185) BAN + c)t) 0, tested by Generalized Method of Moments, *Obs = 51 Estimates of a b and c and their (standard errors) Chi—square Cdt =3) and its significance level) b = 0, c = 408120$ I a+b(s)t—l)—s)t)) +c(sLt—2)—s)t—1)) (1%): month month month 11 22 21 33 22 23 : labi Expctation Formation Extrapolative expectation with one lag (t,k) — s(t) b < b = b > Cases; — s)t)) + (t) belief in a bandwagon belief in constant appreciation distributed lag form a = b = H: bCs(t—1) a + belief in random alk Estimates of a and b and their )standard errors) CHISQ for Hypothesis H: chisg(df=2) and (significance level) AR1 process on is assumed RHO is not reported here HORIZON MONTH a AVE -.015 BAN SEC IRA b —.011 (.002) (.035) 49.42 -.01' (.001) (.044) -.011 —.058 (.003) (.061) 001 —.020 —.008 —.029 (.002) (.068) EXP INS IMP —.009 061 (.001) (.039> —.018 015 (.003) (.067) —.018 -.134 (.003) (.075) FOR CHISO —.001 —.007 (.000) (.007) b> b >0 b < (k) MONTH a month 29 month 10 31 003 006 001 Table 7: Consistency Tests One—month )k2) vs three—month (k=6) expectations s(t,l> st,3 H: — s(t) d + as(t> + bs(t—l> + — s(t) d + As(t> + Bs(t—I) + Cs(t—3.) Consistency d AVE -.0261 0050) BAN SEC -.0061 (.0102) IRA -.0388 (.0093) EXP INS ( 0002 0000 -.0003 —.0272 -.0005 0002) (.0002) ( 0001) ( 0062) (.0001) -.0003 -.0003 —.0056 -.0014 0005 (.0003> (.0003) (.0002) (.0152) (.0004) 0001 (.0002) (GMM) B A 0005 (.0001) ( C 0003 2182.1 (.0001) 300 0000 0005 0001) (.0001) 0001 (.0003) 0009 0003 —.0003 -.0471 -.0007 (.0003) (.0002) (.0073> (.0002) (.0002) H CHISQ 0013 (.0001) 493.0 000 39.7 000 -.0001 12403.3 (.0001) 000 0000 0056) (.0001) 0002 -.0002 —.0142 -.0011 (.0002) (.0001) (.0050) (.0001> 0002 0010 (.0002> (.0001) 77.2 0004 (.0003) 0006 —.0008 —.0423 - 0004 (.0003> (.0002) (.0085) (.0002) 0010 —.0005 (.0002) (.0001) 428 .0006 (.0003) -.0005 -.0000 —.0148 -.0010 (.0004) (.0003) (.0208) (.0002) 0002 0008 3718.9 000 (.0003) (.0002> 0000 —.0000 (.0000> (.0000) 0000 —.0116 (.0000) (.0008) 0000 0001 -.0001 (.0000) (.0000) (.0000) -.0497 —.0346 (.0114) FOR MONTH c 0001 —.0003 —.0254 -.0008 0003 0071) (.0071) (.0001) (.0002) (.0001) -.0190 (.0096> IMP (standard errors> MONTH (OLS) a b —.0202 0061> D=(2+a+b+()1+a)**EH*d A=c1+(2*(l+a)*b>((1a)**3) restrictions: Estimates and cs(t—) —.0053 (.0009) In Micro Data, for only individuals, H is failed to be rejected .000 000 53.6 000 'able )k6> Three—month s(t,3) s(t) — — vs T ele—month (k12) expectations d + s(t) (continued) as(t) d + + As(t) cs(t—ó) bs(t—3) + Bs(t—3) Cs(t—6) + D(2+a)*d Consistency Restrictions: A) )1+a)**2+b)-1 3.) +a) *b+c C( 1+a)*c a AVE H MONTH (3MM) MONTH (OLS) d b 0218 -.0009 000' .0220) (.0003) (.0003) c A D B 0002 (.0002> 0508 -.0019 (.0285) (.0003> 0006 0002) C ) CHIS 0008 0001) 570.1 000 BAN 0201 -.9053 (.0255) (.000'.) 0003 (.0003) 0003 (.0002) 0161 —.0019 (.0388) (.000'.) 0009 (.0003) 0007 (.0002) 326.1 000 SEC 0911 —.0024 (.0306) (.0005) 0011 (.0004) 0005 (.0003) 0988 —.0033 (.0342) (.0005) 0018 (.0003) 0007 (.0002) '.83.8 0105 -.0008 (.0288) (.0004) 0002 0003 (.0003) —.0027 (.000'.) (.0392> (.0005) 0007 (.0003) 0010 (.0002) 607.8 000 TRA EXP INS 0230 -.0006 000' (.0249> (.0004) (.0003) 0001 (.0002> 0865 —.0018 (.0300) (.0003) 0004 (.0002> 0008 (.0002) 211.9 000 0000 —.0003 —.0089 -.0002 0007 (.0003) '.35.9 0003 (.0006> (.0005) 0009 (.0004) 153.0 000 FOR 0002 -.0000 -.0000 (.0001) (.0000> (.0000) 0004 —.0496 (.0294) IMP 131' .000 ) 000'.) (.000'.> ) 0003) 0513 —.0013 0009 —.0000 (.0391> ) 0006) (.0009) (.0004> -.0183 (.0036) 0001 (.0001) -.0000 -.0000 (.0000) ) 0260> -.0005 ) 0003) (.0003) 0112 -.0015 (.0407) —.0344 (.0000> (.0050> In Micro Data, for only individuals, H is failed to be rejected .000 18.6 000 ...NBER Working Paper #2679 August 11988 FOREIGN EXCHANGE RATE EXPECTATIONS: MICRO SURVEY DATA ABSTRACT This conducted paper analyzes the panel data of bi—weekly surveys, by the Japan Center for International... recently that (see, data on foreign exchange rates have become available survey been analyzed and Dominguez (1986) and Frankel and Froot (1987a.b) have exploited the survey data made available... case of micro survey data and fifty-one observations individuals formation at individual j, j individual a j: u(t) is the expected exchange given rate for (3.1) a pure spot exchange rate at time