Asset Price Bubbles: Existence, Persistence and Migration

• Date: 01 March 2017
• Author: Juan P. Franco,Jair N. Ojeda‐Joya,Jose E. Gomez‐Gonzalez,Jhon E. Torres
• DOI: http://doi.org/10.1111/saje.12108
• Published date: 01 March 2017

ASSET PRICE BUBBLES: EXISTENCE, PERSISTENCE

AND MIGRATION

JOSE E. GOMEZ-GONZALEZ

*,

JAIR N. OJEDA-JOYA

†,

JUAN P. FRANCO

‡

AND JHON E. TORRES

§

Abstract

Econometric tests are performed for the detection and migration of asset-price bubbles in the

housing, currency and stock markets of seven countries. This set of countries includes both

developed and emerging economies that have good historical data on housing prices. Our

empirical results suggest that this type of exuberant behaviour in prices occurs more frequently in

the housing market than in the currency and stock markets. Additionally, we find significant

evidence of bubble migration across markets within some of the studied countries.

JEL Classification: G01, G12, C22

Keywords: Financial bubbles, explosive behaviour, housing market, stock market, currency market

1. INTRODUCTION

Efficient markets are supposed to be bubble-free. However, the US housing crash that

gave origin to the recent international financial crisis renewed interest in studying the

existence of bubbles in financial assets. An issue that has gained interest in the literature

is that of identifying bubbles before they burst.

In this work we perform tests for identifying bubbles in asset prices and for

encountering possible migrations between financial assets. We focus in stock market,

currency and housing price bubbles. We apply the detection test proposed by Phillips

et al. (2012), which also allows the date stamping of multiple bubbles in time series. The

tests of bubble migration are based on the work of Phillips and Yu (2011).

Seven countries are studied in this paper. Country selection was based on the

availability of historical data on housing prices with a monthly frequency. The

methodology is implemented with monthly data because it allows a larger number of

observations and allows working with larger window sizes. If quarterly data were chosen,

instead of monthly data, minimum window sizes of 18 periods will require, for bubble

identification purposes, an explosive behaviour of minimum 18 quarters. In other words,

* Corresponding author: Senior Research Economist, Research Unit, Banco de la República.

E-mail: jgomezgo@banrep.gov.co.

†Research Unit, Banco de la República.

‡Maastricht University.

§Inflation Department, Banco de la República

The authors are grateful to Steve Koch and an anonymous referee for their helpful comments and

insights. We also want to thank Hernando Vargas, Angelo Duarte and Esteban Gómez for their

useful discussions, and Oscar Jaulín, Julian Cantor and Carlos Palaciosfor their research assistance.

The errors and omissions are the sole responsibility of the authors. The opinions expressed hereare

those of the authors and do not necessarily represent those of Banco de la República or its Board

of Directors.

South African Journal of Economics Vol. ••:•• •• 2015

© 2015 Economic Society of South Africa. doi: 10.1111/saje.12108

1

V

C2015 Economic Society of South Africa. doi: 10.1111/saje.12108

52

South African Journal of Economics Vol. 85:1 March 2017

South African Journal

of Economics

only bubbles that have duration of at least four and a half years could be identified.

Shorter lived bubbles, as those usually found in stock markets, for example, would be

ruled out due to the low frequency of the data and the requirement of sufficiently large

window sizes for the performance of right-tailed unit root tests. In order to perform unit

root tests, a minimum number of data is required. Using minimum window sizes of less

than 12 observations would imply difficulties for performing unit root tests. If quarterly

data were used instead, this requirement of at least 12 observations for performing a unit

root test would imply that we would rule out bubbles of less than three years of duration.

This would mean that we would be unable to identify bubbles in stock and exchange rate

markets – as well as many housing markets – and would be impossible for us to perform

migration tests.

Phillips et al. (2015) suggest using minimum window sizes using a formula that

depends on the number of observations of the total sample. Following this formula,

our minimum window sizes would vary, depending on the particular country and

market, between 18 and 46. We chose 18 as our minimum window size for two main

reasons. First, using the formula would lead us to having very large minimum window

sizes in some cases. Identifying bubbles in the stock market with periods of at least 40

months of exuberant price behaviour, for instance, would rule out the identification of

bubbles in this market. However, a bubble might have occurred but of shorter

duration. Second, in two related papers that also use monthly data, Yiu et al. (2013)

and Gómez-Gonzalez et al. (2013) use minimum window size of 12 periods, for

practical reasons. Gómez-Gonzalez et al. (2013) show that their results for Colombia

remain valid whenever minimum window sizes of 12, 18 and 24 months are used.

However, if larger minimum window sizes are used, it is hard to identify bubbles in the

housing market.

Phillips et al. (2015) also suggest using a formula to disregard short-lived asset price

bubbles. According to the formula they propose, in our data we should disregard bubbles

of less than 6 months. However,once again we would be unable to identify bubbles in the

stock and exchange rate markets, and it would not be possible to perform migration tests.

Hence, we do not impose a minimum duration time for identifying bubbles.

The selected countries are South Africa, Colombia, the Netherlands, the United

Kingdom, Portugal, South Korea and Canada. Although the US is an interesting country

in the study of financial bubbles and migrations, we do not include it in this study due

to two reasons. First, existence of bubbles in the housing market and several financial

markets has recently been studied for the US. Particularly, Phillips and Yu (2011) use the

same method we use for encountering bubbles in the housing and bond markets.

Meanwhile, Phillips et al. (2012) have done so for the stock market. Additionally,

migration of bubbles between these markets in the US has also been studied by Phillips

and his co-authors. Second, one of the markets we study for the seven countries included

in our sample is the exchange rate market. We consider in all cases the bilateral exchange

rate of each country with respect to the US dollar. Therefore, if the US were included in

our study, we would be unable to study this particular market for this country.

The results of the econometric tests of bubble detection and identification show that

price exuberance is most commonly found in the housing market. Bubbles in currency

and stock markets are less frequent and of shorter duration. These results hold valid for

emerging and developed countries. We find evidence of migration of bubbles betweenthe

asset markets of South Africa, South Korea, the Netherlands and Canada.

South African Journal of Economics Vol. ••:•• •• 20152

© 2015 Economic Society of South Africa.

53South African Journal of Economics Vol. 85:1 March 2017

V

C2015 Economic Society of South Africa.