Skip to main content
Log in

Phase transition in the S&P stock market

  • Regular Article
  • Published:
Journal of Economic Interaction and Coordination Aims and scope Submit manuscript


We analyze the returns of stocks contained in the Standard & Poor’s 500 index from 1987 until 2011. We use covariance matrices of the firms’ returns determined in a time windows of several years. We find that the eigenvector belonging to the leading eigenvalue (the market) exhibits a phase transition. The market is in an ordered state from 1995 to 2005 and in a disordered state after 2005. We can relate this transition to an order parameter derived from the stocks’ beta and the trading volume. This order parameter can also be interpreted within an agent-based model.

This is a preview of subscription content, log in via an institution to check access.

Access this article

Price excludes VAT (USA)
Tax calculation will be finalised during checkout.

Instant access to the full article PDF.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5

Similar content being viewed by others


  1. Extending the window too much could in fact at some point lead to problems, even if one assumes that the betas are slowly varying, see for example Livan et al. (2012).

  2. In this paper \(t\) is used to specific days, while \(\tau \) is used to index variables like \(C\) that are calculated for a time window.

  3. The dominance of the IT sector and the change around 2005 can also be found for smaller window sizes of down to 2 years, revealing one sharp peak around 1995/96. For much smaller time windows the influence of single events like the 1987 stock market crash or the 9/11 attacks become rather large and distort long-run trends.

  4. See any textbook on quantum mechanics, e.g., Messiah (1967).


  • Ahlgren N, Antell J (2010) Stock market linkages and financial contagion: a cobreaking analysis. Q Rev Econ Finance 50:157–166

    Article  Google Scholar 

  • Alfarano S, Lux T, Wagner F (2005) Estimation of agent-based models: the case of an asymmetric herding model. Comp Econ 26:19–49

    Article  Google Scholar 

  • Adcock CJ, Cortez MC, Rocha Armada MJ, Silva F (2012) Time varying betas and the unconditional distribution of asset returns. Quant Finance 12(6):951–967

    Article  Google Scholar 

  • Beine M, Candelon B (2011) Liberalization and stock market co-movement between emerging economies. Quant Finance 12(2):299312

    Google Scholar 

  • Black F, Jensen MC, Scholes M (1972) The capital asset pricing model: some empirical tests In: Jensen M (ed) Studies in the theory of capital markets. Praeger, New York

  • Bos T, Newbold P (1984) An empirical investigation of the possibility of stochastic systemic risk in the market model. J Bus 57(1.1):35–41

    Article  Google Scholar 

  • Bodurtha JN, Mark NC (1991) Testing the CAPM with time-varying risk and returns. J Finance 46(4):1485–1505

  • Bollerslev T, Engle R, Wooldridge J (1988) A capital asset pricing model with time-varying covariances. J Polit Econ 96(1):116–131

  • Borland L, Hassid Y (2010) Market panic on different time-scales. arXiv:1010.4917v1

  • Bornholdt S (2001) Expectation bubbles in a spin model of markets. Int J Mod Phys C 12:667–674

    Article  Google Scholar 

  • Brock WA, Hommes CH (1998) Heterogeneous beliefs and routes to chaos in a simple asset pricing model. J Econ Dyn Control 22:12351274

    Article  Google Scholar 

  • Burda Z, Görlich A, Jarosz A, Jurkiewicz J (2004) Signal and noise in correlation matrix. Phys A 343:295

    Article  Google Scholar 

  • Chiarella C, Dieci R, He X (2010) A framework for CAPM with heterogeneous beliefs. In: Bischi G-I, Chiarella C, Gardini L (eds) Nonlinear dynamics in economics, finance and social sciences: essays in honour of John Barkley Rosser Jr. Springer, Berlin

    Google Scholar 

  • Citeau P, Potters M, Bouchaud JP (2001) Correlation structure of extreme stock returns. Quant Finance 1:217–222

    Article  Google Scholar 

  • Cont R, Bouchaud JP (2000) Herd behaviour and aggregate fluctuations in financial markets. Macroecon Dyn 4:170

    Article  Google Scholar 

  • Cont R (2001) Empirical properties of asset returns: stylized facts and statistical issues. Quant Finance 1:223

    Article  Google Scholar 

  • Dichev ID, Huang K, Zhou D (2012) The dark side of trading. Emory law and economics research paper No. 11–95

  • Fama EF, French KR (1992) The cross-section of expected stock returns. J Finance 47(2):427–465

    Article  Google Scholar 

  • Galluccio S, Bouchaud JP, Potters M (1998) Rational decisions, random matrices and spin glasses. Phys A 259:449–456

    Article  Google Scholar 

  • Harvey CJ, Siddique A (2000) Conditional skewness in asset pricing tests. J Finance LV(3):1263–1295

    Article  Google Scholar 

  • Kenett DY, Shapira Y, Madi A, Bransburg-Zabary S, Gur-Gershgoren G, Ben-Jacob E (2011) Index cohesive force analysis reveals that the US market became prone to systemic collapses since 2002. PLoS ONE 6(4):e19378

    Article  Google Scholar 

  • Kenett DY, Preis T, Gur-Gershgoren G, Ben-Jacob E (2012a) Dependency network and node influence: application to the study of financial markets. Int J Bifurc Chaos 22. doi:10.1142/S0218127412501817

  • Kenett DY, Raddant M, Lux T, Ben-Jacob E (2012b) Evolvement of uniformity and volatility in the stressed global financial village. PLoS ONE 7(2):e31144

    Article  Google Scholar 

  • Kirman A (1993) Ants, rationality and recruitment. Q J Econ 108:137–156

    Article  Google Scholar 

  • Laloux L, Cizeau P, Bouchaud JP, Potters M (1999) Noise dressing of financial correlation matrices. Phys Rev Lett 83:1467

    Article  Google Scholar 

  • Lintner J (1965) The valuation of risk assets and the selection of risky investments in stock portfolios and capital budgets. Rev Econ Stat 47:13–37

    Article  Google Scholar 

  • Livan G, Alfarano S, Scalas E (2011) Fine structure of spectral properties for random correlation matrices. Phys Rev E 84:016113

    Article  Google Scholar 

  • Livan G, Inoue J, Scalas E (2012) On the non-stationarity of financial time series: impact on optimal portfolio selection. JSTAT. doi:10.1088/1742-5468/2012/07/P07025

  • Lux T, Ausloos M (2002) Market fluctuations I: scaling, multi-scaling and their possible origins. In: Bunde A, Kopp J, Schellhuber H (eds) Theory of desaster. Springer, Berlin, p 373

    Google Scholar 

  • Marĉenko VA, Pastur LA (1967) Distribution of eigenvalues for some sets of random matrices. Sb Math 1:457

    Article  Google Scholar 

  • Messiah A (1967) Quantum mechanics volume I. North Holland, Amsterdam

    Google Scholar 

  • Münnix MC, Shimada T, Schäfer R, Leyvraz F, Seligman T, Guhr T, Stanley HE (2012) Identifying states of a financial market. Sci Rep 2(644).

  • Pearson K (1885) Notes on regression and inheritance in the case of two parents. Proc R Soc Lond 58(240):242

    Google Scholar 

  • Plerou V et al (2002) Random matrix approach to cross correlations in financial data. Phys Rev E 65:066126

    Article  Google Scholar 

  • Preis T, Schneider JJ, Stanley HE (2011) Switching processes in financial markets. Proc Natl Acad Sci. doi:10.1073/pnas.1019484108

  • Sharpe W (1964) Capital asset prices: a theory of market equilibrium under conditions of risk. J Finance 19:425–442

    Google Scholar 

  • Stauffer D, Sornette D (1999) Self-organized percolation model for stock market fluctuation. Phys A 271:496–506

    Article  Google Scholar 

  • Tumminello M, Aste T, Di Matteo T, Mantegna RN (2005) A tool for filtering information in complex systems. Proc Natl Acad Sci USA 102(30):10421–10426

    Article  Google Scholar 

  • Wagner F (2006) Application of Zhangs square root law and herding to financial markets. Phys A 364:369–384

    Article  Google Scholar 

  • Wishart J (1928) The generalized product moment distribution in samples from a normal multivariate population. Biometrica 20A(1928):32–52

    Article  Google Scholar 

  • Wu FY (1982) The Potts model. Rev Mod Phys 54(1):235268

    Article  Google Scholar 

Download references

Author information

Authors and Affiliations


Corresponding author

Correspondence to Matthias Raddant.


Appendix 1: Correlation matrix

Denoting the time average as in Eq. (3) by \([\;]_{\tau ,t_W}\) we get from Eqs. (4) and (5) the average of \(r_M^2\)

$$\begin{aligned}{}[r_M^2]_{\tau ,t_W}=\frac{\lambda _0(\tau )}{N} \end{aligned}$$

Similarity we get for \([r_{av}^2]_{\tau ,t_W}\) and \([r_M\cdot r_{av}]_{\tau ,t_W}\)

$$\begin{aligned}{}[r_{av}^2]_{\tau ,t_W}=\frac{1}{N}\sum _{\mu =0}^{N-1}a_\mu ^2(\tau )\lambda _\mu (\tau ) \end{aligned}$$


$$\begin{aligned} a_\mu (\tau )=\frac{1}{\sqrt{N}}\sum _ie_i^\mu (\tau ) \end{aligned}$$

\(a_0\) corresponds to the mean value \(\bar{\beta }\) of \(\beta _i\) over \(i\).

$$\begin{aligned}{}[r_M\cdot r_{av}]_{\tau ,t_W}=\frac{\lambda _0(\tau )}{N}\cdot \bar{\beta } \end{aligned}$$

Inserting Eqs. (24), (25) and (27) into \(\varDelta ^2\) from (4) we get

$$\begin{aligned} \varDelta ^2=(1-\bar{\beta })^2 + \sum _{\mu >0}\frac{\lambda _\mu (\tau )}{\lambda _0(\tau )}a_\mu ^2(\tau ) \end{aligned}$$

Since \(e_i^\mu \) and \(e_i^0\) are orthogonal for \(\mu >0\) we can write \(a_\mu \) as

$$\begin{aligned} a_\mu (\tau )=\frac{1}{\sqrt{N}}\sum _i e_i^\mu (1-\sqrt{N}e_i^0) \end{aligned}$$

Applying the Schwartz inequality to (29) we get

$$\begin{aligned} a_\mu ^2(\tau )\le 2(1-\bar{\beta }) \end{aligned}$$

Together with \(\sum _{\mu =0}\lambda _\mu =trace(C)\) this leads to the inequality (13) for \(\varDelta ^2\). Since the average \(C_{av}\) is a function \([r_{av}^2]_t\) insertion of \(\varDelta ^2\) into Eq. (25) leads to Eq. (15).

Appendix 2: Perturbation expansion

The matrix \(C\) in Eq. (17) is a sum of two matrices. The first \(C^0_{ij}=\beta ^0_i\beta ^0_j\gamma _M^2\) has one large eigenvalue \(E_0=\gamma _M^2N\) with an eigenvector \(f^0_i=\beta ^0_i/{\sqrt{N}}\) and \(N-1\) degenerate zero eigenvalues with vectors \(f^\mu _i\) with \(\mu >0\).These must satisfy only the orthogonality relation

$$\begin{aligned} (f^0,f^\mu )=0 \end{aligned}$$

with (\(a,b)\) denoting the scalar product. To obtain a complete basis we impose on \(f^\mu \) in the \(N-1\) dimensional subspace the following conditions with the second matrix \(C^1_{ij}=\delta _{ij}\gamma _i^2\)

$$\begin{aligned} (f^\nu ,C^1\; f^\mu )=0 \quad \text{ for } \; \mu \ne \nu \quad \text{ and }\; \mu , \nu \ne \; 0 \end{aligned}$$

We apply standard second order Rayleigh Schrödinger perturbation theoryFootnote 4 with \(C^1\) as perturbation. For general matrices \(C^0\) and \(C^1\) using only the spectrum of \(C^0\) and condition (31) we get up to order \(1/E_0^2\) for the leading eigenvalue

$$\begin{aligned} \lambda _0=E_0+(f^0,C^1f^0)+\frac{1}{E_0}[(f^0,(C^1)^2f^0) -(f^0,C^1f^0)^2] \end{aligned}$$

and its eigenvector

$$\begin{aligned} e^0_i=f^0_i+\frac{1}{E_0}[(C^1f^0)_i-(f^0,C^1f^0)f^0_i] \end{aligned}$$

The other eigenvalues require the in general complicated solution of Eq. (17) for \(f^\mu _j\). They are given by

$$\begin{aligned} \lambda _\nu =(f^\nu ,C^1\;f^\nu )-\frac{1}{E_0}(f^0,C^1f^\nu )^2 \end{aligned}$$

Inserting the specific form of \(C^1\) we obtain for \(\lambda _0\) and \(\beta _i\) with

$$\begin{aligned} (f^0,C^1f^0)= & {} \frac{1}{N}\sum _i(\beta ^0_i)^2\gamma _i^2=\langle \gamma ^2\rangle _\beta \end{aligned}$$
$$\begin{aligned} (f^0,(C^1)^2f^0)= & {} \frac{1}{N}\sum _i(\beta ^0_i)^4\gamma _i^2=\langle \gamma ^4\rangle _\beta \end{aligned}$$
$$\begin{aligned} \lambda _0= & {} \gamma _M^2N+\langle \gamma ^2 \rangle _\beta +\frac{1}{\gamma _M^2N} \left[ \langle \gamma ^4 \rangle _\beta -(\langle \gamma ^2 \rangle _\beta )^2 \right] \end{aligned}$$
$$\begin{aligned} \beta _i= & {} \beta ^0_i \left( 1+\frac{1}{\gamma _M^2N} \left( \gamma ^2_i-\langle \gamma ^2 \rangle \right) \right) \end{aligned}$$

\(\langle \;\rangle _\beta \) denotes the average over \(i\) weighted with \((\beta ^0_i)^2\). Since neglected terms are of order \(1/N^2\) these formulae describe \(\lambda _0\) and \(\beta _i\) fairly accurate already for moderate \(N\). Due to the degeneracy the general formalism of Marĉenko and Pastur (1967) for the modification due to noise does not apply. Using the Wishart (1928) formula we find the relative error in \(\lambda _0\) due to a finite observation window \(T\) is of order \(1/\sqrt{T}\) instead of \(\sqrt{N/T}\) expected from Marĉenko and Pastur (1967). The non-leading eigenvalue will be changed considerably if the spread of \(\gamma _i\) is small, see Burda et al. (2004).

Appendix 3: Significance of the order parameter

We analyzed the significance of the order parameter by assigning a random value for the sector to each stock. In Fig. 6 we show one representative result of such a random assignment. Compared with Fig. 5, the large values of \(m\) disappear. Before and after 2006 \(m\) behaves similar as in a disordered phase. The structure observed near 2006 cannot be removed by re-shuffling, because a substantial part of the stocks change their values of \(\beta _i v_i\). If there would be only one stock in each IT or financial sector responsible for the transition seen in Fig. 5, one would get the same behavior. The fluctuation size of \(m\) in the order of \(\pm 0.15\) indicates the error on the estimation of \(m\). This shows that the observed values for the IT and financial sector are far out of the range of random fluctuations.

Fig. 6
figure 6

The risk parameter \(m(\tau ,s)\) with the same \(\beta _i\) and \(v_i\) as in Eq. (20) but for shuffled sector affiliations \(s\)

Rights and permissions

Reprints and permissions

About this article

Check for updates. Verify currency and authenticity via CrossMark

Cite this article

Raddant, M., Wagner, F. Phase transition in the S&P stock market. J Econ Interact Coord 11, 229–246 (2016).

Download citation

  • Received:

  • Accepted:

  • Published:

  • Issue Date:

  • DOI: