Skip to main content
Log in

When bubbles burst: econometric tests based on structural breaks

  • Regular Article
  • Published:
Statistical Papers Aims and scope Submit manuscript


Speculative bubbles have played an important role ever since in financial economics. During an ongoing bubble it is relevant for investors and policy-makers to know whether the bubble continues to grow or whether it is already collapsing. Prices are typically well approximated by a random walk in absence of bubbles, while periods of bubbles are characterised by explosive price paths. In this paper we first propose a conventional Chow-type testing procedure for a structural break from an explosive to a random walk regime. It is shown that under the null hypothesis of a mildly explosive process a suitably modified Chow-type statistic possesses a standard normal limiting distribution. Second, a monitoring procedure based on the CUSUM statistic is suggested. It timely indicates such a structural change. Asymptotic results are derived and small-sample properties are studied via Monte Carlo simulations. Finally, two empirical applications illustrate the merits and limitations of our suggested procedures.

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

Access this article

Subscribe and save

Springer+ Basic
EUR 32.99 /Month
  • Get 10 units per month
  • Download Article/Chapter or Ebook
  • 1 Unit = 1 Article or 1 Chapter
  • Cancel anytime
Subscribe now

Buy Now

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

Instant access to the full article PDF.

Fig. 1
Fig. 2

Similar content being viewed by others


  1. Phillips et al. (2011c) suggest critical values that slowly tend to infinity in order to ensure that the break dates are estimated consistently. In finite samples, however, the distribution of estimated break dates is severely skewed leading to a considerable delay of 5–15 periods [cf. Homm and Breitung (2010)].

  2. Note that on exchange markets the stock price immediately adjusts for dividends so that the price represents the ex dividend price of the share. It follows that in periods with dividend payment the ex dividend price also have the martingale difference property.

  3. To simplify the exposition we neglect a possible constant or time trend in the autoregressive representation. In our empirical examples a constant is included.

  4. As pointed out by Ploberger and Krämer (1992), the forecast residuals can be replaced by the ordinary (in-sample) residuals. In the explosive model, however, the distribution of the cumulated residuals depends on the unknown parameter \(c\). We therefore do not consider the OLS CUSUM approach in what follows.

  5. We also tried out various other rules such as the ones suggested by Schwert (1989). Overall the square-root rule performs best in our simulations.

  6. Please note that this estimate differs from the one reported in Phillips et al. (2011c) as we consider a sub-sample running from June 1993 to June 2003.

  7. Note that there is an obvious typo in Theorem 4.3(a) of Phillips and Magdalinos (2007), cf. their Eq. (9).


  • Bachelier L (1900) Théorie de la Spéculation, Annales Scientifiques de l’Ecole Normale Supérieure, 3rd. 17:21–88. (Translated in: The random character of stock market prices, edited by Paul Cootner (1964), Cambridge)

  • Bai J (1994) Least squares estimation of a shift in linear processes. J Time Ser Anal 15:453–472

    Article  MathSciNet  MATH  Google Scholar 

  • Blanchard OJ, Watson MW (1982) Bubbles, rational expectations, and financial markets. In: Wachtel P (ed) Crisis in the economic and financial structure. Lexington Books, Lexington, pp 295–315

  • Chow GC (1960) Tests of equality between sets of coefficients in two linear regressions. Econometrica 52:211–222

    Google Scholar 

  • Chu C-SJ, Stinchcombe M, White H (1996) Monitoring structural change. Econometrica 64:1045–1065

    Article  MATH  Google Scholar 

  • Davis RA, Dunsmuir WTM (1996) Maximum likelihood estimation for MA(1) processes with a root on or near the unit circle. Econ Theory 12:1–29

    Article  MathSciNet  Google Scholar 

  • Froot K, Obstfeld M (1991) Intrinsic bubbles: the case of stock prices. Am Econ Rev 81:1189–1214

    Google Scholar 

  • Homm U, Breitung J (2010) Testing for speculative bubbles in stock markets: a comparison of alternative methods. University of Bonn, Bonn. Working Paper

  • Homm U, Breitung J (2012) Testing for speculative bubbles in stock markets: a comparison of alternative methods. J Financ Econ 10(1):198–231

    Google Scholar 

  • Krämer W, Sonnberger H (1986) The linear regression model under test. Physica-Verlag, Heidelberg

  • Kruse R, Frömmel M (2012) Testing for a rational bubble under long memory. Quant Financ 12:1723–1732

    Article  Google Scholar 

  • Newey WK, West KD (1987) A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica 55:703–708

    Article  MathSciNet  MATH  Google Scholar 

  • Phillips PCB, Magdalinos T (2007) Limit theory for moderate deviations from a unit root. J Econ 136:115–130

    Article  MathSciNet  Google Scholar 

  • Phillips PCB, Yu J (2011) Dating the timeline of financial bubbles during the subprime crisis. Quant Econ 2:455–491

    Article  MathSciNet  MATH  Google Scholar 

  • Phillips PCB, Shi S, Yu J (2011a) Testing for multiple bubbles. Working Paper, Research collection school of economics. Paper 1302.

  • Phillips PCB, Shi S, Yu J (2011b) Specification sensitivity in right-tailed unit root testing for explosive behavior. Sim Kee Boon Institute for Financial Economics, Singapore Management University, Working Paper

  • Phillips PCB, Wu Y, Yu J (2011c) Explosive behavior in the 1990s Nasdaq: when did exuberance escalate asset values? Int Econ Rev 52:201–226

    Article  Google Scholar 

  • Ploberger W, Krämer W (1992) The CUSUM-test with OLS-residuals. Econometrica 60:271–286

    Article  MathSciNet  MATH  Google Scholar 

  • Schwert W (1989) Test for unit roots: a Monte Carlo investigation. J Bus Econ Stat 7:147–159

    Google Scholar 

Download references


We would like to thank an anonymous referee and a Guest Editor for their helpful comments. Robinson Kruse gratefully acknowledges financial support from CREATES funded by the Danish National Research Foundation.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Jörg Breitung.

Appendix: Proof of Theorem 1

Appendix: Proof of Theorem 1

(i) Under the null hypothesis we have

$$\begin{aligned} \sum \limits _{t=[\tau T]+1}^T \tilde{e}_t^0 y_{t-1}&= \sum \limits _{t=[\tau T]+1}^T \varepsilon _t y_{t-1} + (\varrho - \widehat{\varrho }^0) \sum \limits _{t=[\tau T]+1}^T y_{t-1}^2 \\&= \sum \limits _{t=[\tau T]+1}^T \varepsilon _t y_{t-1} - \left(\frac{ \sum \nolimits _{t=2}^T \varepsilon _t y_{t-1} }{ \sum \nolimits _{t=2}^T y_{t-1}^2 } \right) \ \sum _{t=[\tau T]+1}^T y_{t-1}^2 \\&= \sum \limits _{t=[\tau T]+1}^T \varepsilon _t y_{t-1} - \left( \frac{ \sum \nolimits _{t=[\tau T]+1}^T y_{t-1}^2 }{ \sum \nolimits _{t=2}^T y_{t-1}^2 }\right) \ \sum \limits _{t=2}^T y_{t-1}\varepsilon _t \\&= \sum \limits _{t=[\tau T]+1}^T \varepsilon _t y_{t-1} - \kappa _T \sum \limits _{t=2}^T y_{t-1}\varepsilon _t, \end{aligned}$$


$$\begin{aligned} \kappa _T = \frac{ \sum \nolimits _{[\tau T]+1}^T y_{t-1}^2 }{ \sum \nolimits _{t=2}^T y_{t-1}^2 }. \end{aligned}$$

We now show that \(\kappa _T=1 - o_p(1)\). From Phillips and Magdalinos (2007) we have as \(T\rightarrow \infty \)

$$\begin{aligned} \sum _{t=2}^T y_{t-1}^2 = O_p\left(\varrho ^{2T}T^{2\gamma } \right) \end{aligned}$$

and, thus,

$$\begin{aligned} \kappa _T&= 1- \frac{\sum \nolimits _{t=2}^{[\tau T]} y_{t-1}^2 }{ \sum \nolimits _{t=2}^T y_{t-1}^2 } = 1-\frac{ O_p( \varrho ^{2\tau T}(\tau T)^{2\gamma }) }{ O_p( \varrho ^{2 T} T^{2\gamma }) } ~=~ 1- O_p\left[\tau ^{2\gamma } \left( \varrho ^{-2(1-\tau ) T} \right) \right]. \end{aligned}$$


$$\begin{aligned} \lim _{T\rightarrow \infty } \varrho ^{-2(1-\tau ) T}&= \lim _{T\rightarrow \infty } \left[ \left( 1+\frac{c}{T^\gamma } \right)^{ T^\gamma /c} \right] ^{ -2(1-\tau )cT/{T^{\gamma }} } \\&= \lim _{T\rightarrow \infty } e^{ -2c(1-\tau )T^{1-\gamma }} \rightarrow 0 \end{aligned}$$

for any \(\gamma <1\), it follows that \(\kappa _T \stackrel{p}{\rightarrow } 1\).

In a similar manner it can be shown that

$$\begin{aligned} \frac{ \sum \nolimits _{[\tau T]+1}^T \varepsilon _t y_{t-1} }{ \sum \nolimits _{t=2}^T \varepsilon _t y_{t-1} } = 1 + o_p(1). \end{aligned}$$

Therefore, \(\sum \nolimits _{t=[\tau T]+1}^T \varepsilon _t y_{t-1}\) is \(O_p\left(\varrho ^{T}T^{\gamma } \right)\) (cf. Phillips and Magdalinos (2007), Theorem 4.3a). It follows that

$$\begin{aligned} \frac{\varrho ^{-\tau T}}{(\tau T)^{\gamma }} \sum _{t=[\tau T]+1}^T \tilde{e}_t^0 y_{t-1} = - \frac{\varrho ^{-\tau T}}{(\tau T)^{\gamma }} \sum _{t=2}^{[\tau T]} \varepsilon _t y_{t-1} + o_p(1). \end{aligned}$$

Using \((\tau T)^{-1} \sum _{t=2}^{[\tau T]} (\tilde{e}_t^0)^2 \stackrel{p}{\rightarrow } \sigma _u^2\) we obtain

$$\begin{aligned} \psi _\tau&= \frac{ \sum \nolimits _{t=[\tau T]+1}^T \tilde{e}_t^0 \, y_{t-1} }{ \sigma _u \sqrt{ \sum \nolimits _{t=[\tau T]+1}^T y_{t-1}^2 } } + o_p(1) \\&= \tau ^{\gamma } \left( \varrho ^{-(1-\tau ) T} \right) \frac{\frac{\varrho ^{-\tau T}}{(\tau T)^{\gamma }} \sum \nolimits _{t=[\tau T]+1}^T \tilde{e}_t^0 \, y_{t-1} }{ \sigma _u \sqrt{\frac{\varrho ^{-2T}}{T^{2\gamma }} \sum \nolimits _{t=[\tau T]+1}^T y_{t-1}^2 } } + o_p(1) \\&= o_p(1), \end{aligned}$$

and, therefore, \(\psi _\tau \) converges to zero in probability as \(T\rightarrow \infty \) and \(c>0\).

(ii) From Theorem 4.3 of Phillips and Magdalinos (2007) it follows thatFootnote 7

$$\begin{aligned} \frac{\varrho ^{-\tau T}}{(\tau T)^{\gamma }} \sum _{t=2}^{[\tau T]} \varepsilon _t y_{t-1}&\Rightarrow XY \\ \frac{\varrho ^{-2\tau T}}{(\tau T)^{2\gamma }} \sum _{t=2}^{[\tau T]} y_{t-1}^2&\Rightarrow \frac{1}{2c} Y^2 \end{aligned}$$

for all \(0<\tau <1\), where \(X\) and \(Y\) are independent \(\mathcal{N}(0,\sigma _\varepsilon ^2/2c)\) random variables. Thus,

$$\begin{aligned} \psi _\tau ^*&= \frac{\frac{\varrho ^{-\tau T}}{(\tau T)^{\gamma }} \sum \nolimits _{t=[\tau T]+1}^T \tilde{e}_t^0 \, y_{t-1} }{ \sigma _\varepsilon \sqrt{\frac{\varrho ^{-2\tau T}}{(\tau T)^{2\gamma }} \sum \nolimits _{t=[\tau T]+1}^T y_{t-1}^2 } } + o_p(1) \\&\Rightarrow \frac{ XY }{\frac{\sigma _\varepsilon }{\sqrt{2c}} |Y|} = \frac{\sqrt{2c}}{\sigma _\varepsilon } \ \text{ sign}(Y)X \sim \mathcal{N}(0,1). \qquad \diamond \end{aligned}$$

Rights and permissions

Reprints and permissions

About this article

Cite this article

Breitung, J., Kruse, R. When bubbles burst: econometric tests based on structural breaks. Stat Papers 54, 911–930 (2013).

Download citation

  • Received:

  • Revised:

  • Published:

  • Issue Date:

  • DOI:


JEL Classification