First, we create an algorithm that classifies bonds into different levels of liquidity, using appropriate liquidity measures and two different methods (SNNs and logistic regression). This procedure provides a ground for a comparison of how well either approach captures relevant liquidity characteristics, which, in turn, enables the researchers to determine whether the increased complexity of the SNN provides any improvements in liquidity classification compared to a basic logistic regression.
Part II: Self-Normalizing Neural Networks - Bond Liquidity Classification
In the second part of the article series, we outline a framework utilising both the Self-Normalizing Neural Networks (SNNs) and the logistic regression for bond liquidity classification. This framework is subsequently applied to the Swedish bond market in an investigative case study.