Not logged in.

Contribution Details

Type Master's Thesis
Scope Discipline-based scholarship
Title Financial Market Anomalies: Acceleration Effect and Gamma Factor
Organization Unit
Authors
  • Simona Ferrari
Supervisors
  • Marc Paolella
Language
  • English
Institution University of Zurich
Faculty Faculty of Business, Economics and Informatics
Number of Pages 135
Date 2018
Zusammenfassung Over the last years, there has been increasing criticism of fundamental asset pricing labels, such as the Capital Asset Pricing Model (CAPM) or its extended version Fama-French Three-Factor Model as well as of theories such as the Efficient Market Hypothesis (Avramov & Chordia, 2006). Behavioral economists and other scientists have argued that due to a process of simplification – based on a number of strong assumptions such as the predictability and rationality of investor behavior - those asset pricing models are unable to explain numerous financial market anomalies which seem to stem from the behavioral irrationality of the investor (Fama, 1998). There is evidence that overconfidence, overreaction as well as other behavioral biases and risk-based sources are the cause of one of the most important asset pricing anomalies: “momentum” (Blitz, Hanauer, & Vidojevic, 2017) (Daniel, Hirshleifer, & Subrahmanyam, 1998). According to Jegadeesh & Titman (1993): “if stock prices either overreact or underreact to information, then profitable trading strategies that select stocks based on their past returns exist”. Indeed, the two most popular trading investment styles are the “contrarian” by De Bondt & Thaler (1985) and “momentum” documented for the first time by Jegadeesh & Titman (1993), which aim to achieve benefits from the “mean-reversion effect” (for the former) and the “short-term return persistence” anomalies (for the latter). Momentum was first defined and documented in 1993 in a paper by Jegadeesh & Titman (1993); it was described by Fama & French (1993) as “the premier unexplained anomaly”. Momentum consists in the persistence of a linear trend in the log-price process and there is empirical evidence across countries and time as well as asset classes (Fama & French, 1993). Nowadays, momentum-based strategies, i.e. strategies based on the D (delta) factor are widely implemented by asset managers. Recently, Ardila, Forr`o, & Sornette (2015) reported the evidence of an important novel effect complementing momentum: “acceleration”, which is defined as the change in momentum and is quantified by “the first difference of successive returns”, i.e. by the gamma (G) parameter. Ardila, Forr`o, & Sornette (2015) defined acceleration as “transient (non-sustainable)” phenomenon related “to positive feedbacks influencing the price formation, which is prevalent during “special market regimes”. Furthermore, it emerged that acceleration is related with procyclical mechanisms such as psychological and behavioral aspects. The study revealed that, on average, G-allocations have a positive performance and according to different parametrizations outperform momentum-based strategies in about two out of three cases. This research is an extension of the previous paper by Ardila, Forr`o, & Sornette (2015) and it aims firstly to develop better proxies to detect the D and the G parameter; successively two investment strategies (the Long-Short and the Relative StrengthWeighted Portfolio) are optimized according to the D or the G factor to investigate portfolio performance. Three kind of detection methodologies are implemented to quantify the momentum and the acceleration effect: the simple, the trend-based and the wavelet transform (i.e. the Maximum Overlap Discrete Wavelet Transform, MODWT) approach. By applying the simple approach, the D and the G parameter are quantified as in the paper by Ardila, Forr`o, & Sornette (2015), i.e. momentum is defined as the f -months cumulative return while acceleration is measured as the f -months difference in momentum. Moreover, since according to previous literature, “momentum” is defined as a short/medium-term persistence in log-returns, the trend-based detection aims to Page III University of Zurich, Empirical Finance, September 21, 2018 improve the detection of the D factor using time series analysis tools as moving averages. Hence, a first trend-based approach implements an Exponential Moving Average to extrapolate the time series trend (i.e. momentum) whilst removing irregular fluctuations and noises. Moreover, a second trend-based detection estimates the trend as the difference between a short and a long Simple Moving Average. Thereafter, acceleration is computed as in the simple approach. According to Lera & Sornette (2017) and Shao & Ma (2003) the n-derivative of a signal is given by its convolution with a wavelet having nvanishing moments. Since the momentum effect (D) might also be represented as the “velocity” of stock prices, it can be quantified by the first derivative of the log-price time series. Moreover, acceleration (G) might be defined (as in physics) as a change in velocity and it can be modelled by the second derivative. Hence, according to previous literature, in order to detect the delta factor, a wavelet transform using a Daubechies function with one vanishing moment (also named Haar wavelet function) is applied. Furthermore, the acceleration factor is captured by the detail coefficients of a MODWT performed with a Daubechies function with two vanishing moments (Db2). Moreover, the investigation is also executed using winsorized data. Finally, a new hybrid portfolio optimization strategy which aims to consider both the momentum and the acceleration effect as factors for optimization has been developed, namely the D/G (Delta-Gamma) optimization. This strategy is an extension to the ”traditional” time-series momentum strategy. More precisely, by applying the D/G optimization, a long and a short portfolio are constructed selecting stocks according to two conditions: the direction of momentum (“delta condition”) and the direction of acceleration (“gamma condition”); moreover, equal-weights or relative G-weights are applied. More precisely, the long portfolio buys stocks with a positive momentum (i.e. a positive D) and having an upward accelerating price (i.e. a positive G), both factors are detected over the same formation period or at the same resolution level. Moreover, the short portfolio sells stocks having a negative momentum (i.e. a negative D) and a downward accelerating price (i.e. a negative G ). The investigation has been performed considering the U.S. equity market over two different periods in time: the distant past (1984-2002) and the recent past (2001-2016). This study adds convincing evidence about the lower or even negative performance of momentum (as well as the acceleration) strategies, during the recent past (2001-2016), a period of time characterized by the dramatic impact of the global financial crisis (2007-2009) and therefore by a more volatile financial market regime. Previous literature on ”momentum crashes” under unstable and stressed market states is sufficient to explain this outcome. Furthermore, on average, an improved portfolio performance is possible using D and G factors detected through the trend-based as well as the Maximum Overlap Discrete Wavelet Transform approach. Additionally, an important contribution is given by the newly developed hybrid D/G strategy. Indeed, there is significant evidence that implementing the hybrid portfolio optimization, i.e. a more ”flexible” but more ”selective” investment strategy which considers both the momentum and the acceleration as factors for optimization and which does not invest in a constant number of assets allows us to even gain good returns during stressed and more volatile market regimes. Page
PDF File Download
Export BibTeX