This action might not be possible to undo. Are you sure you want to continue?

arXiv:cond-mat/9803374v3 [cond-mat.stat-mech] 11 May 1998

**Parameswaran Gopikrishnan, Martin Meyer, Lu´ A. Nunes Amaral, ıs and H. Eugene Stanley
**

Center for Polymer Studies and Department of Physics, Boston University Boston, MA 02215, U.S.A. (February 1, 2008)

Abstract

The probability distribution of stock price changes is studied by analyzing a database (the Trades and Quotes Database) documenting every trade for all stocks in three major US stock markets, for the two year period Jan 1994 – Dec 1995. A sample of 40 million data points is extracted, which is substantially larger than studied hitherto. We ﬁnd an asymptotic power-law behavior for the cumulative distribution with an exponent α ≈ 3, well outside the Levy regime (0 < α < 2). PACS numbers: 89.90.+n

Typeset using REVTEX 1

. 2 .84 ± 0. the New York Stock Exchange (NYSE).The asymptotic behavior of the increment distribution of economic indices has long been a topic of avid interest [1–6]. P (g) ≡ P {gi(t) ≥ g}. Here. the probability for an increment larger or equal to a threshold g. and . gi (t) ≡ [Gi (t) − Gi (t) ] /vi . as a function of g. since the asymptotic behavior can be obtained only by a proper sampling of the tails.03 (positive tail) and α = 2. The basic quantity of our study is the relative price change. the share price multiplied with the number of outstanding shares). Figure 1a shows the cumulative probability distribution.000 normalized increments gi (t) per company per year.12 (negative tail) from two to hundred standard deviations. diﬃcult to obtain. where the volatility vi ≡ (2) (Gi (t) − Gi(t) )2 of company i is measured by the standard deviation. i = 1 . We obtain about 20. which requires a huge quantity of data. 1.e. however. which is much larger than studied hitherto. We normalize the increments. we analyze a database documenting each and every trade in the three major US stock markets. . 1995 [7]. the American Stock Exchange (AMEX).e. where Si is the market price of company i (i. i.1 ± 0. Gi (t) ≡ ln Si (t + ∆t) − ln Si (t) ≃ Si (t + ∆t) − Si (t) . 1000 is the rank of the company according to its market price on Jan. We form 1000 time series Si (t). . is a time average [8]. Conclusive empirical results are. which gives about 4 × 107 increments for the 1000 largest companies in the time period studied. 1994. and the National Association of Securities Dealers Automated Quotation (NASDAQ) for the entire 2-year period Jan. . We thereby extract a sample of approximately 4 × 107 data points. 1994 to Dec. The data are well ﬁt by a power law P (g) ∼ g −α (3) with exponents α = 3. Si (t) (1) where the time lag is ∆t = 5 min.

such as vi ≡ |Gi(t) − Gi(t) | . Acknowledgements: We thank X. Figure 1b shows the results for the negative and positive tail respectively. Mantegna.73 ± 0. we perform several additional calculations: (i) we change the time increment in steps of 5 min up to 120 min. The results are all consistent with α = 3. 3 . Peng and D. Stauﬀer for helpful discussions. ’96 using the same methods as above (Fig.12 for the positive and α = 2. To test the robustness of the inverse cubic law α ≈ 3. Liu.84 ± 0. (ii) a L´vy distribution [2.In order to test this result. We test this method by analyzing two surrogate data sets with known asymptotic behavior. where the tails become “approximately exponential” [3]. and unlike (i) and (ii) it is not a stable distribution. each using all increments larger than 5 standard deviations. S. To put these results in the context of previous work. The e inverse cubic law diﬀers from all three proposals: Unlike (i) and (iii). we calculate the inverse of the local logarithmic slope of P (g). These extensions will be discussed in detail elsewhere [10]. Gabaix. The method yields the correct results 3 and ∞ respectively.K. we recall that proposals for P (g) have included (i) a Gaussian distribution [1]. (iii) we replace deﬁnition of the volatility by other measures. 1c and d). it has diverging higher moments (larger than 3). We estimate the asymptotic slope α by extrapolating γ as a function of 1/g to 0. (a) an independent random variable with P (x) = (1 + x)−3 and (b) an independent random variable with P (x) = exp(−x). (ii) we analyze the S&P500 index over the 13y-period Jan.12]. C.13 for the negative tail. R. and (iii) a e truncated L´vy distribution. Havlin. γ −1 (g) ≡ −d log P (g)/d log g [9].11. and DFG and NSF for ﬁnancial support. ’84 – Dec. Extrapolation of the linear regression lines yield α = 2. Y.

Potters. Th´orie de l’addition des variables al´atoires (Gauthier-Villars. 1937). 3. Peinke. Mandelbrot. Dodge. A. Ecole Norm. 1163 (1975). Empirical Finance 3. Pagan. J. Sup. W. Bachelier. [7] The Trades and Quotes Database. [4] S. Amaral. Theorie des Risques Financieres. Eyrolles 1998). Europ. J. J. Stat. N. and Y. Mantegna and H. Ghashgaie. 1897). (Alea-Saclay. E. published by the New York Stock Exchange. 46 (1995). Paris. 24 CD-ROM for ’94-’95. Ann. Meyer. Gopikrishnan. L. Stanley. to be published. 767 (1996). (cond-mat/9705075). 15 (1996). 21 (1900). Talkner. J. ´ [11] V. Sci.REFERENCES ´ [1] L. Nature 376. Ann. Breymann. M. [12] P. Cours d’Economie Politique (Lausanne and Paris. [9] B. [6] J. E. Hill. [8] In order to obtain uncorrelated estimators for the volatility and the price increment at time t. and references therein. 294 (1963).P. [10] P. Pareto. Cont. M. Nature 381. N. B (1998). and H. Stanley. R. B. [3] R. 3. P. e e e 4 . Phys. Bouchaud and M. Business 36. the time average extends only over time steps t′ = t. [5] A. in press. L´vy. [2] B.

2 0.05 0.2 0..20 1/g FIG.02 for the positive and negative tails respectively. except that the number of increments per data point is 100. g(1) > g(2) > . The regression lines yield α = 2.10 0. The lines in (a) are power law ﬁts to the data over the range 2 < g < 100. (c) Same as (a) for the 1 min S&P500 increments. tail neg.4 0. γ(g) ≡ − (d log P (g)/d log g)−1 as a function of 1/g for the negative (◦) and the positive (+) tail respectively. fit γ(g) exponential 10 −8 1 10 100 g 10 0 0.3 0.93 and α = 3.0 Levy (lower bound) / b P(g) 10 −4 10 −6 neg.10 0.05 0.00 0. > g(N ) . fit pos.. fit pos.0 P(g) 10 −4 10 −6 neg. 1. fit 1 10 100 γ(g) exponential 10 −8 g 0.20 1/g c 0. 5 . tail neg.5 a 10 −2 0.3 0. by sorting the normalized increments by their size. (b) the inverse local slope of P (g). We obtain an estimator for γ(g). and we obtain for the local slope γ(g(k) ) = − log(g(k+1) /g(k) )/ log(P (g(k+1) /P (g(k) )) ≃ k(log(g(k+1) ) − log(g(k) )).00 0.FIGURES 10 0 0.15 0. Note that the average m−1 m (k) ) k=1 γ(g over all events used would be identical to the estimator for the asymptotic slope proposed by Hill [9].1 0. (a) Log-log plot of the cumulative probability distribution P (g) of the normalized price increments gi (t).5 Levy (lower bound) / d 10 −2 0.15 0. tail pos. Each data point shown in b is an average over 1000 increments g(k) . (d) Same as (b) for the 1 min S&P500 increments.1 0. The cumulative density can then be written as P (g(k) ) = k/N . and the lines are linear regression ﬁts to the data. tail pos.4 0.

- Handy Book on Cbs
- KottaPanchangam Free KinigeDotCom
- Alternative Energy
- DevGuru_JetSQL_Quickref
- Automatic Fib Levels
- Pascal Tick
- Rajan Met a Stock
- Kanakadhara Stothram Telugu PDF
- Market Profile
- Values
- Pacal Effective Volume
- TDI_1
- Gann
- ForumFaqsk tios
- Sindhu Ra
- Renko Ashi Trading System 2
- Super Trend
- Heiken Ashi
- Pacal Effective Volume
- Values
- 004 Dave
- 63975642-Add-Ins
- 63975642-Add-Ins
- EV page 261
- MFLib DLL API Documentation

Sign up to vote on this title

UsefulNot useful- CourseWork1 20122013 Solutions
- ch6
- Bayes Tutorial2
- Introduction to Logarithms
- UWEETR-2010-0002
- lecture3.ppt
- Funcion Exponencial IB Respuestas
- 0lecture_05
- 012 CAT 2003 ReTest Sol
- 13041_1990-1994
- Spring07 OBrien T
- Manual Logarithms
- The Tick-By-tick Dynamical Consistency of Price Impact in Limit Order Books
- Chapter 1-Review of algebra.pdf
- q Sar Lecture
- Std09 III Maths Em
- Eviews Demonstration
- 06_tanal
- Tutorial of EM Algorithm
- Standard Form, Indices and Logarithm
- Class11_-_2011
- PHAR2821-QSAR-lec1-150805
- History of logarithms
- How to Use Log Tables
- OWASP Logging Guide
- 2011 Discovery Round 3 Solutions
- math
- Basic Scientific Maths
- Mathematics Methods Sample Calc Free Marking Key 2016
- Woon1998_FractalZeta_9812031
- Inverse Cubic

Are you sure?

This action might not be possible to undo. Are you sure you want to continue?

We've moved you to where you read on your other device.

Get the full title to continue

Get the full title to continue reading from where you left off, or restart the preview.

scribd