A Long‐Run Performance Perspective on the Technology Bubble

• Published date: 01 May 2018
• Date: 01 May 2018
• DOI: http://doi.org/10.1111/fire.12162

The Financial Review 53 (2018) 379–412

A Long-Run Performance Perspective

on the Technology Bubble

Maximilian Franke and Gunter L¨

offler∗

Ulm University

Abstract

The events surrounding the stock price peak of March 2000 are commonly interpreted as

the bursting of a technology or Internet bubble, with some researchers pointing out that the

pattern could also arise in fundamental models. Weinform the debate by studying the long-run

performance of Internet and technology stocks from March 2000 onward. Using calendar-time

regressions, we do not find conclusive evidence of negative abnormal returns. The results are

consistent with a new interpretation of the events; namely, the price drop of the early 2000s

was not warranted in light of future cash flows and risk.

Keywords: Internet stocks, tech stocks, bubble, long-run abnormal performance

JEL Classifications: G10, G12, G14

1. Introduction

Many Internet and technology firms appear to have been overvalued in the late

1990s. For example, Amazon’s stock price fell from $66.75 in March 2000, which

∗Corresponding author: Institute of Finance, Ulm University,Helmholtzstraße 18, 89081 Ulm, Germany;

Phone: +49 (0)731/50-23597; Fax: +49 (0)731/50-23950; E-mail: gunter.loeffler@uni-ulm.de.

This research was supported by the German Research Foundation through the Collaborative Research

Center 649 “Economic Risk,” Humboldt-Universit¨

at zu Berlin. We are very grateful to an anonymous

referee for numerous valuablecomments. We would also like to thank Robert Resutek and other conference

participants of the 65th Annual Meeting of the Midwest Finance Association 2016 for their comments and

suggestions.

C2018 The Eastern Finance Association 379

380 M. Franke and G. L¨

offler/The Financial Review 53 (2018) 379–412

at the time was difficult to rationalize with fundamental models (see Damodaran,

2000), to $5.91 in October 2001. At the time of this writing, in August 2017, however,

Amazon is trading above $950. Should we re-evaluate the interpretation of the 2000–

2001 meltdown, in light of the long-run performance that is now observable to us?

This is the question we want to address in this paper—not for Amazon, but for the

entire Internet and technology segment.

The mainstream view is that the year 2000 saw the bursting of a bubble.1

Researchers support this view by pointing to extraordinarily high valuation levels

(e.g., Ofek and Richardson, 2002), irrational investorbehavior (e.g., Cooper, Dimitrov

and Rau, 2001), and price explosiveness (e.g., Phillips, Wu and Yu, 2011; Phillips,

Shi and Yu, 2015). On the other hand, Pastor and Veronesi (2006, 2009), who enrich

standard valuation models through uncertainty and changes in risk characteristics,

conclude that both the high price level and the subsequent drop can be rationalized.

To illustrate the difficulty of reaching a conclusion from the analysis of prices

and valuation levels at the peak, we apply the bubble dating approach of Phillips, Shi

and Yu (2015) to the NASDAQ composite. Essentially, this method uses data until

a date tto test whether prices at tcame about through a run-up that appears to be

nonstationary. Results are shown in Figure 1.

The late 1990s clearly stand out by showing exceptional, explosive behavior.

However, other periods show exceptional behavior as well, even though they are

usually not classified as bubble periods, such as the years 2013 and 2015. More

importantly, the analysis only indicates that something exceptional was going on; it

does not establish that prices were too high compared to fundamentals. In the words

of Phillips, Shi and Yu (2015, p. 1046), exuberance visible in the time series “may

be rational or irrational depending on possible links to market fundamentals.”

By studying long-run performance, we can shed light on the question of whether

the valuation levelsat the peak were consistent with fundamentals or not. Specifically,

we examine the followinghypothesis: the price level of Internet and technology stocks

at the peak was fair, whereas the price drop was too large. Once the pricing errors

have been corrected, we should not find average abnormal returns. The hypothesis is

consistent with the observation of Fama (2014) that many large stock price declines

are quickly compensated wholly or in large part by price increases. This observation

leads to the question if price drops could also be driven by irrational pessimism.

Siegel (2003) suggests a similar approach as we do but, at the time, does not yet

have long-run performance data; he also favors a pragmatic performance evaluation.

How best to control for risk is not evident. Severalmethodological problems in testing

long-run abnormal returns are familiar from the literature (Barber and Lyon, 1997;

Lyon, Barber and Tsai, 1999; Mitchell and Stafford, 2000). In our study, some prob-

lems are potentially more severe due to a high number of delistings and potentially

1For the purpose of this paper, a bubble can be regarded as a price levelthat is too high to be justified by

fundamental valuation. It can include both rational and irrational bubbles.

M. Franke and G. L¨

offler/The Financial Review 53 (2018) 379–412 381

Figure 1

Price explosiveness in the NASDAQindex

The figure plots the results from applying the date stamping strategy of Phillips, Shi and Yu(2015) (lower

panel) to the deflated time series of the NASDAQcomposite index (upper panel) from January 1971 until

February 2017. To locate price explosiveness or collapse, the backward sup augmented Dickey–Fuller

test statistic is compared to its critical value, which is obtained from Monte Carlo simulations with 2,000

replications. The solid line shows the backward sup augmented Dickey–Fuller test statistics, the dashed

line the corresponding critical values, and the dotted line the deflated NASDAQ starting at 100 in January

1971. We use the consumer price indexm04128 from the Federal Reserve Bank of St. Louis. The shaded

areas mark periods in which the test indicates price explosiveness.

large changes in firm characteristics. That is why we restrict the approaches to test for

long-run abnormal returns to calendar-time regressions; namely, the Bessembinder

and Zhang (2013) approach and factor model regressions. Bessembinder and Zhang

(2013) capture time-varying risk through explanatory variables in abnormal return

regressions, while factor model regressions can deal with it through wandering factor

sensitivities. In contrast, time series variation in the characteristics of sample and

control firms is not captured in the buy-and-hold abnormal return approach. Further-

more, calendar-time regressions offer the advantage over the buy-and-holdabnormal