Detection, modelling and implications of non-normality in financial economics : normal inverse Gaussian modelling of Norwegian stock market returns and consumption growth
Master thesis
Permanent lenke
http://hdl.handle.net/11250/302166Utgivelsesdato
2015Metadata
Vis full innførselSamlinger
- Master Thesis [4487]
Sammendrag
This thesis shows that the Norwegian stock market deviates significantly from what one might
think of as a baseline model with identically and independently normally distributed returns.
Firstly, the stock market return does not seem to be normally distributed over any observation
frequency (daily, monthly and quarterly) we have investigated in this thesis. More specifically,
the return distribution is both leptokurtic and negatively skewed. Secondly, the empirical return
distribution is time-varying; we find both autocorrelation in returns and volatility clustering. Both
of these deviations from the baseline model can potentially have important implications for
theoretical models and practical applications.
In this paper, we will model the return distribution with a normal inverse Gaussian (NIG)
distribution, which we indeed find to outperform Gaussian distributions both in- and out of
sample. Our NIG modelling approach allows us to deviate from the normality assumption, but it
is not able to capture the dependencies across time. This model of returns turns out to be useful in
risk measurement, where the baseline model grossly underestimate well-known metrics such as
value at risk and expected shortfall the NIG model fits these measures nicely.
This thesis also applies a bivariate NIG distribution to a theoretical model of equilibrium risk-free
interest rates and the equity premium, suggested by Aase and Lillestøl (2015), in order to explain
the equity premium puzzle. The NIG model allows for fatter tails and negative skewness in the
joint return and consumption distribution, thereby reducing the implied risk aversion parameter
and increasing the impatience rate of the representative consumer. Although the model takes us in
the right direction in terms of both implied parameters, the improvement is only slightly more
than negligible and it happens at the cost of a great increase in complexity.