Initial public offerings: an asset allocation decision based on nonnormal returns

Abstract

This study analyses whether the inclusion of initial public offering (IPO) stocks as part of an optimal asset allocation strategy can reduce the systematic risk of an investment portfolio. The asset allocation framework takes account of nonnormality of returns from a universe of eight asset classes using 240 monthly returns between January 1992 and December 2011. AR-GJR-GARCH models resolve the tail dependence and heteroscedasticity in the return series. Generalized Pareto distributions help to fit the heavy-tailed return distributions, while copula functions help to calibrate the dependence structure between the asset returns. The optimization algorithm persistently includes IPO stocks as part of an optimal asset allocation strategy. Their portfolio inclusion reduces the conditional value-at-risk and improves the risk-return trade-off.

Keywords:

• initial public offerings
• strategic asset allocation
• generalized pareto distribution
• copula

JEL Classification:

• C14
• C32
• C63
• G11

Acknowledgements

I am grateful to Josef Schuster for providing the data on the US initial public offering index IPOX100. IPOX^® is a registered international trademark of IPOX^® Schuster LLC; US Patent No. 7,698,197. I thank the editor, an anonymous referee and Alistair Bruce for helpful comments.

Notes

1 ¹Other performance tracking IPO indexes include, for example, Schuster's IPOX Global 100, IPOX Global 30 All Markets, IPOX Global 30 Developed Markets, IPOX China 20, IPOX30 Asia-Pacific, IPOX Europe 50 and IPOX30 Europe.

2 ²The IPOX100 is a dynamically, quarterly reconstituted, value-weighted index. Index membership is fixed to the top 100 IPO stocks, measured by the firm's market capitalization. Therefore, the IPOX100 represents a fully diversified portfolio of IPO stocks, which minimizes any concerns in relation to unsystematic risk. More detailed information on the IPOX100 is available from http://www.ipoxschuster.com. Alternative IPO-tracking indexes are also available. However, these indexes have comparatively shorter historical price data available for analysis and the absolute number of constituents is smaller than that of the IPOX100.

3 ³Brooks et al. (2005) construct semi-parametric distributions using GPD for the tails and kernel estimates for the interior of the distributions. This approach provides superior results to other methods. Gupta and Liang (2005) estimate the capital adequacy of hedge funds through the value-at-risk approach using GPD. Hotta et al. (2008) use ARMA-GARCH models to estimate the joint return distribution of portfolios from the innovations using copula functions. In their study, they model the marginal distributions with the help of GPD. McNeil and Frey (2000) combine GARCH models and EVT to estimate the conditional value-at-risk and the conditional expected shortfall of returns. This approach outperforms the simple square-root-of-time scaling method. Zhao et al. (2010) combine GARCH models with EVT and GPD to provide accurate estimates of the lower and upper tails of index return distributions.

4 ⁴Glasserman et al. (2002) use t-copulas to model portfolio risk of heavy-tailed distributions. Hotta et al. (2008) apply conditional copulas to construct joint distributions for a two-asset portfolio. Rosenberg and Schuermann (2006) employ copulas to aggregate market, credit and operational risk.

5 ⁵Some of the key events include: Japanese asset price bubble (1991–2003), Asian financial crisis (1997), Russian financial crisis (1998), dot-com bubble (2000–2001), economic effects from the 9/11 attacks (2001), stock downturn (2002), Chinese stock bubble (2007), financial crisis (2007–2009), Dubai debt standstill (2009), sovereign debt crisis (2010–2011) and stock markets fall (2011).