Вы находитесь на странице: 1из 6
Economics Letters 46 (1998) 427-482 7 (65-1765 /94/ $0700 © 1984 Elsevier Science BLY. All rights reserved Should we care? Psychological barriers in stock markets Kees G. Koedijk Department of Finance, University of Limburg, P.O. Box 616, 6200 MD Maasrict, Netherlands Philip A. Stork** Department of Intemational Economics Erasmus University, P.O. Box 1738, 3000 DR Rowerdam, Netherlands Reveived 8 April 1993, Accepted 30 August 1993 Abstract For thre out of five major stockmarket it is found tht those levels which are 1 multiple ofa hundred are approsched and transgressed relatively infrequently. We explicily test for the effect of sample bias and conduct 2 forecasting experiment. JEL clasifcation: G14 1. Introduction The existence of so-called psychological barriers in financial markets is a relatively un- documented anomaly. Only recently have Donaldson (1990) and De Grauwe and DeCupere (1992) succeeded in shedding some more light on this subject. They showed that there is clear evidence of certain levels in both stock markets and foreign exchange markets to which market participants attribute more significance than to other levels. Such levels appear not to be based on fundamental economic theory, but probably are created and kept alive by mass psychology. In stock markets these barriers often are associated with those levels which are a multiple of a hundred or a thousand. Newspapers, for instance, often report statements like: Today, the Nikkei stock index dropped through the important 20,000 resistance level.” Apparently, approaching these levels bears with it several consequences for the behaviour of market participants as revealed in stock prices, In the current study we employ the concept of psychological barriers in stock markets as introduced into finance by Donaldson (1990). To this purpose we use daily middle rates from Datastream (average of bid and ask rates) between 1 January 1980 and 28 February 1992 (i.e. 3174 * Corresponding author " We are grateful to Frank Kleibergen, Clemens Kool, Theo Nijman, Timo Malta, Alber de Vaal and Casper de Vies for comments and suggestions on a previous version of this paper. Any remaining errors ae of couse our own. The second author gratefully acknowledges grant $80-229-006 of NWO, the Dutch Research Foundation SSDI 0165-1765(93)00343-M. 28 K.G. Koedik, P.A. Stork Economies Leters 44 (1994) 427-432 observations) for the Brussel Stock Exchange (Belgium), the FAZ General (Germany), the Nikkei Stock Average 225 (Japan) and the Standard and Poors Composite (United States). Also wwe use the FTSE-100 (United Kingdom) between 2 January 1984 and 28 February 1992, i.e, 2130 observations. In section 2 we apply several tests to examine the existence of positional effects in the indices of five major stock markets. The set-up of our tests is more precise than the tests in Donaldson (1990) or De Grauwe and DeCupere (1992) and allows for a more detailed analysis of the existence of certain effects. Moreover, we test explicitly for the possibility of sample bias by considering several (sub) samples. In section 3 we use a similar approach to examine the existence of transgressional effects. In section 4 we try to answer the obvious question: Can these psychological barriers be exploited profitably? Or stated differently, should we care about the existence of certain effects of psychological barriers in stock markets? Hence, in section 4 we investigate whether the information on psychological barriers can be used to predict returns in stock markets. We conclude in section 5 by summarizing the main results. 2. Positional effects All levels in the stock markets for which the last two digits are zero are regarded as potential psychological barriers. Hence, for the German Faz index, for instance, which hovered between 200 and 800 during the period under consideration, we regard the levels 200, 300. ..800 as potential psychological barriers. Therefore, for each index only the last two digits are used, after rounding it to an integer. The values of these integers range betwcen 00 and 99 and shall henceforth be referred to as the M-values of the index. Also, a vector P(M) with length 100 is created, which registers the relative frequency of each M-value occurring, If there are no psychological barriers, we would expect each M-Value to have approximately the same amount of occurrences and to be distributed uniformly. In order to test this we divide the Mcvalues in ten disjunct categories of equal size, i.e. 06-15, 16-25. . 96-05. For each category we register the number of times the index closes with an M-value inside this category. A chi-squared goodness-oF fittest is used to compare the actual and hypothetical number of observations per category. The test-statistic A and the significance level for which the uniformity of the M-values is rejected are reported in Table 1 "The results show that for four out of five stock market indices the M-values are not uniformly distributed. We test, by means of a simple regression test, whether this apparent non-uniform distribution of the M-values is linked to the presence of psychological barriers. Three dummies are introduced with value 1 if the index is within a certain distance of a psychological bartier and zero otherwise. The dummies D,, D, and D, equal 1 if the M-value of the index is in one of the following ranges: 98, 99,00, 1,02 (for’D,); 93,...,97 or 3,...,7 (for D3); 85,..-,98 or 8,... 15 (for D,). We regress the P(M) vector on a constant C and the dummy vectors: P(M)=C+B,D, + B.D; + BsDs a Table 1 Chisquared test for positional effects Brussel FAZ FISE Nikkei SeP a 79 89 1698 tae 1722 Sign. level 0.00% 0.00% 497% 25.67% 0.00% K.G. Koedik, P.A. Stork 1 Ecomomice Leuers 44 (1994) 427-432 9 Table 2 Regression test for positional effects Bruel FAZ, FISE, Nikkei SaP c 106 iis Lo 100 ian 829) (7.0) (8. 2s) (72) a 0.26 -0st -o11 =0.06 070 (2565) ar 12 (035) (505) 8, -013 -0sr 008 0.06 -0.48 cm (3.82) 068) (098) 43 & 0.20 046, 005 0.04 -oat (316) 47) (0.84) om 121 Hence, we examine if the index halts relatively (in)frequently close to a 00-level by computing the B-estimates. If these are negative, then apparently there are relatively few observations close to these O0-levels. The results in Table 2 confirm that this is indeed the case. The first number in each row is the parameter estimate of either C or Band the second number (between brackets) is the f-value of a significance test for that parameter against zero. Nearly all B-estimates are negative, but not significantly so for the FTSE and the Nikkei indices. The fact, that the estimates of f, tend to decrease if / increases reflects that farther away from the psychological barriers the effect diminishes. For the Brussel index, for instance, the relative frequency of the last two digits between 16 and 84 is 1.06%, as opposed to 1.00% if the M-values ‘were distributed uniformly. The frequency of the last two digits with M-values 98, 99, 00, 01 or 02 is 1.06% —0.26% = 0.80%, distinctly less than expected in the case of a uniform distribution. ' We checked for the robustness of the existence of the psychological barriers by splitting up the full sample period into four sub-samples: 1980-1982, 1983-1985, 1986-1988 and 1989-1991. For the FTSE index we consider only the last two sub-samples. The chi-squared test and the regression test are conducted for those four separate sub-samples. We do not report all the results here because of lack of space, but restrict ourselves to the main conclusions. * The positional effects are present in the FAZ, the FTSE and the S&P indices for most of the sub-samples. The positional effects for the Brussel index, however, are now insignificant for three sub-samples, with very significant effects in the first period only. The results for the Nikkei index remain insignificant. All in all, the positional effects appear to hold up fairly well for three out of the five indices, in the sense that they are relatively robust to splitting the sample, 3. Transgressional effects In this section we investigate whether the so-called psychological barriers are transgressed less frequently than other levels. For instance, if the index jumps from 235 to 247, the M-values 36 up The set-up of Eq, (1) with three distinctive dummies fr the three areas around the psychological barviers appears to be preferable wo the se-up used by Donaldson (1990) or DeGrauwe and Decupere (1992), who employ one single dummy nd simply expand the area for which tis Gummy equals one, The new area then is a combination ofthe old area and the fide part, and is much harder to interpret. We feel thatthe set-up with three dummies provides more insight in the hehaviou of the index around the pryeological barriers These reslts are avilable from the authors upon reset +0 K.G. Koedijk, P.A. Stork { Economics Leters 44 (1994) 427-492 ‘Table 3 Regression test for transgressional effects Brussel FAZ FISE, Nikkei SAP ¢ ore 108 10 12 (207.22) (2182) es.) (653.10) (3.92) B oat 037 -on -0.00 0.88 (0.68) (9.41) (2.8) (200) (158) Be ~0.03 038 0.0 -0.0 051 250) 1058) (4.44) 1.9) 15) a 0.06 01s -0.0 0.0 0.16 (519) (599) (5.48) (292) (2.66) to and including 47 are transgressed. The vector T(M) records the relative frequency with which each M-value is transgressed. If there are no transgressional effects, all elements of the T(M)- vector should be roughly equal. When we run a simple chi-squared test as in section 2, we find that uniformity of the distribution of transgressed M-values is rejected at the 1% significance level for four out of the five indices (not Nikkei). Whether this is caused by the presence of psychological barriers or not is tested by means of a regression test. The vector T(M) and the three dummies are used in a similar fashion as in the foregoing section. The results are depicted in Table 3. The format of Table 3 is identical to that of Table 2. The negative B parameters show that the indices transgress the O(-levels and their direct neighbourhood relatively infrequently. The strongest effects are found in the FAZ and S&P indices, where the relative frequencies of transgressed M-values 98, 99,00,01 and 02 are 0.71% (1.08~0.37) and 0.44% (1.12~0.68), respectively. For the Nikkei index, the 6, estimates are significantly different from zero, but too small to make a strong case for transgressional effects of psychological barriers. As in the previous section, we checked for the robustness of these effects by splitting up the full sample period. The transgressional effects remain present in the FAZ, the FISE and the S&P indices for most of the sub-samples. The positional effects for the Brussel, index, however, now are insignificant for two sub-samples, with significant negative fs inthe fist period and significant positive B's in the third period. The results for the Nikkei index remain insignificant. Allin all, the transgressional effects appear to hold up fairly well for three out of the five indices, in the sense that they are relatively robust to spliting the sample. 4. Predictability of stock returns ‘A natural extension of the above results is to investigate if the presence of psychological barriers can be used to predict stock returns, The reader should note that predictability does not necessarily imply market inefficiency.’ In order to test whether the presence of psychological barriers leads to predictability, we have tested the following specification over several subperiods: Tu =€+6,D} + 8,D) +8,D} + Br, , Q) "See Hawawini and Keim (1992) for an overview of predictability of stock rturns, or Fama (1991) fora discussion of financial markets efficiency. K.G. Koedijk, PA. Stork | Economies Leners 44 (1994) 427-432 a Table 4 Predictability test FAZ FISE S&P Nikkei Brasel © 0.0582 ‘o.on0a83 0.000824 ‘omn7i 0-000250 (175) as) 46) (22) ast) 9.000486 0.002740 0.001680 0.000376 0.000720 (0335) 2.60) C116) (054) Go o.000218 6.000185 .000224 0.000196 1.000835 02) (02 (026) (031) (1.06) 4 0.000425 6.000883 6.00064 0.000632 0.000229 (085) (073) (1) 2.16) (0.54) 6 0.012209 oaniees 0.035880 noasesi 0.218995 (069) G3) 20) es) (1248) where and where 7, stands for the stock return and s, for the index at time f. The results for the five stock ‘markets are reported in Table 4. The parameter estimates for the dummies of the psychological barriers are not significant. Only in the case of the FTSE do we find some weak evidence of predictability, but this result did not hold over sub-samples. Generally, we can conclude that there is no evidence of predictability of stock returns induced by the presence of psychological barriers. 5. Conclusion In this paper we have presented an empirical analysis of the presence of psychological barriers in five major stock markets using daily data between January 1980 and February 1992. We found that round numbers are approached as well as transgressed relatively infrequently. This finding appeared to hold over several sub-samples and thus proved to be a ‘real’ phenomenon and does not just reflect sample bias. We also tested whether the presence of psychological barriers implies predictability of stock returns. We found that the presence of psychological barriers does not lead to predictability. With respect to the question in the introduction of this paper, ‘Should we care about psychological barriers in stock markets?", the answer is both yes and no. Yes, because psychological barriers are not just a reflection of sample bias but are real. No, because the presence of psychological barriers does not imply predictability of stock returns References De Grauwe, P. and D. Decupere, 199, Psychological barriers i the foreign exchange markets, Journal of International ‘omparative Economics 9, 87-10) 3 K.G. Koedijk, P.A. Stok { Economics Leners 44 (1994) 427-432 Donaldson, R.G., 1990, International evidence on psychological bares in asset prices and the elfiient market hypothesis, Woodrow Wilson School, Princeton University, Financial Research Center Memorandum, No. 116 Fama, E-F., 1991, Efcient capital markets I, Journal of Fiance, 46, 1575-1617 Hawavini, G. and D.B. Keim, 192, On the preditbilty of common stock returns: World-wide evidence forthcoming in Handbook of Finance

Вам также может понравиться