An evaluation of value-at-risk with GARCH vis-a-vis value-at-risk with EWMA or EVT in the valuation of haircuts: Evidence from the repo market in the Philippines

• Main Author: Cao, Harold C.
• Format: text
• Language: English
• Published: Animo Repository 2019
• Subjects: Repurchase agreements > Philippines; GARCH model; Exponentially weighted moving average; Extreme value theory; Finance and Financial Management
• Online Access: https://animorepository.dlsu.edu.ph/etd_masteral/7157
• Institution: De La Salle University

The domestic repo market has been on and off. There were already rules and regulations in place for the domestic repo trading, however, the said rules and regulations must further be developed, strengthened and tested, particularly on the use of haircuts in repo trading. Currently, the domestic repo market participants use ten percent (10%) as the standard haircut rate for all repo transactions, which is aligned with the standards of the Global Master Repurchase Agreement (GMRA). This study developed a new haircut valuation model that is based on the Value-at- Risk (VaR) measure by combining the principles behind the Exponentially Weighted Moving Average (EWMA) and Extreme Value Theory (EVT), which aimed to provide alternative options to the domestic repo market participants in the use of haircut rates in their repo transactions. The study aimed to find a better model between the new haircut valuation model – combination of EWMA and EVT – and the Generalized Autoregressive Conditional Heteroskedasticity (GARCH) model. EWMA and GARCH are two models that are proven to be effective in predicting changes in volatility while the EVT is proven to be a good tool in modelling extreme events and stressed scenarios. The F-test methodology was conducted to examine the variability of the haircut rates generated by the two models. It was concluded that both of the two models can either be used in the valuation of domestic repo haircut rates. GARCH appeared to be a better model in terms of having low variability across the results for most of the benchmark tenor buckets. The EWMA was noted to have great sensitivity to the chosen value of the λ. The new model emerged to be a more effective model during stressed scenarios because of the use of the EVT as well as based on some certain specific scenarios.