Browsing by Author "Lundell, Niklas"

Volatility affects the pricing of many financial instruments and generally acts as a proxy for the risk of an asset. In addition to volatility itself, the persistence of volatility has also been found to play a role in the financial markets. Thus, to be able to accurately model volatility and its persistence is nothing short of important. One of the most common ways to model volatility is to use the GARCH model, which might overestimate the persistence of volatility if the unconditional variance of the time series experiences structural shifts at some point. Several algorithms have been proposed to identify these structural shifts, and many of these utilize the cumulative sum of squares of the series. According to previous empirical findings, the volatility persistence of stock and sector indices and currency exchange rates decreases when structural shifts are incorporated into the GARCH models. These change points also seem to occur roughly at the same time, suggesting that they might have some common social, political or economic cause. Many authors have questioned whether the algorithms used are suitable for financial data, which is often found to be heteroskedastic and leptokurtic. To be more precise, algorithms that do not take into account these properties of financial data may actually find spurious structural shifts in the unconditional variance. One of the focuses of this thesis lies on comparing the performance of two cumulative sum of squares algorithms, of which one can be argued to better suit financial data. Few previous studies have used European data ranging well beyond the 2008 financial crisis. In this thesis, the two algorithms are applied to the stock market index and seven sector indices from the Helsinki stock exchange. The period of observations ranges from 2002 to 2015. The algorithms are used to obtain the structural change points from these indices, after which the change points are incorporated into the GARCH framework. This thesis also examines whether volatility persistence decreases or not for this data, and whether the structural shifts obtained with the algorithms occur simultaneously or not. The finding of this thesis are mostly in line with previous research. The use of the more appropriate algorithm for financial data leads to a significant decrease in the amount of structural shifts detected. When using the most robust methods, only one structural shift is detected in all the eight indices: this is somewhat less than in most previous studies. For this one structural shift, the persistence of volatility decreases as expected, but for the possibly spurious structural shifts, the results are more ambiguous. A striking share of the structural shifts, spurious or not, occur during the end of 2007 and the beginning of 2008. The author of this thesis suggests that future research on the cumulative sum of squares algorithms would benefit from concentrating on how empirical results are affected by sample size and the decision of whether to use daily or weekly data.