IPOs, regime switching and optimal portfolio allocation
Abstract
This thesis studied whether including the information contained in the first day IPO returns
can be used to improve forecasts of the performance of financial markets. The S&P 500
index was used as a proxy for the financial market. The thesis utilized Hamilton’s (1989)
regime switching model in identifying the regimes in the training data. All the models contained
two regimes, representing bull and bear markets. The first model was a regime switching
(RS) model on the S&P 500 log returns, in the second model the VIX and the TED spread
were added as regressors, in the third model the first day IPO returns was also added as a
regressor. A GARCH(1,1) model was also be fitted to the S&P 500 log returns. The RS
models and the GARCH model’s performance was compared on pseudo out-of-sample
performance during the first quarter of 2020. In addition, each RS model were back tested
using Markowitz (1952) portfolio optimization theory, and compared to a portfolio created by
an AR(1) model on the S&P 500 log returns that did not allow for regime switching. The back
testing was done in the time period extending from January 1990 until December 2019. The
results of the study showed that IPO returns had a non-statistically significant positive
correlation with the S&P 500 log returns. In the negative returns regime, the IPO returns
showed the highest level of significance, with a p-value of 0.1915. The results of the pseudo
out-of-sample forecasts showed that the model that included the IPO returns performed the
worst when forecasting the S&P 500 log returns during the first quarter of 2020. The best
performing model in forecasting was the RS model that included the VIX and the TED spread
as regressors, which even outperformed the GARCH model. The GARCH model
outperformed the other RS models in the pseudo out-of-forecast. In back testing, there was a
different story. Here, the best performing model was the RS model that included the IPO
returns, which marginally outperformed the S&P 500 RS model. All the RS models
outperformed the model that did not allow for regime switching. The RS model that included
IPO returns had a holding period return that was 0.3% higher than the S&P 500 RS model.
This comparison was done without considering the costs associated with trading. The S&P
500 RS model required rebalancing 9 times during the period, whereas the RS model that
included the IPO returns required rebalancing 48 times in the same time period. These results
indicate that adding IPO returns to the models did not add value to either the pseudo out-ofsample
forecasts or in portfolio optimization.