The purpose of this paper is to create a "realistic" and "quantitative" model for determining an insurance company's asset allocation. It can also be applied to non-insurance companies with huge balance sheets. It is often difficult to apply a quantitative model to insurance company when one wants to construct such a realistic asset allocation because of following three features.

First (1), the assets of insurance companies are enormous, often exceeding hundreds of billions of dollars, and it is difficult to reorganize it frequently. Second (2), many insurance companies have long-term liabilities, and the risk of them must be considered in risk management. Third (3), insurance companies are often subject to accounting and regulatory constraints.

In this paper, the author have constructed a new model that clears up the above characteristics. In constructing the model, the hierarchical risk parity approach and the Black-Litterman method were used as references. Five main characteristics are presented below. The first point (ⅰ) is that to ensure robustness, the clustering and hierarchical asset allocation optimization by hierarchy is performed by machine learning as seen in the hierarchical risk parity approach. Second (ⅱ), the optimization method incorporates the concept of implied expected return found in the Black-Litterman method. The third (ⅲ) is the addition of constraints on asset changes. The fourth (ⅳ) is that the optimization is limited to assets only and incorporates liability information only for the risk penalty. The fifth (ⅴ) point is the use of a risk-adjusted return measure as the objective function during optimization.

Each of the characteristics corresponds to the insurance company-specific issues raised above, and (ⅰ), (ⅱ), (ⅲ) characteristics of models correspond to the first (1) issue, and (ⅳ) to (2), and (ⅴ) to (3).

The author performed Monte Carlo simulations in Python to compare the new model with existing models to confirm its usefulness.

Find the Q&A here: Q&A on 'Some Novel Investment Approaches'