Description: This article summarizes key points from the recently published research paper “Deep Learning for Liability-Driven Investment,” which applies reinforcement learning and deep learning techniques to liability-driven investment (LDI).

Description: A discussion about how many models can address the over fitting issues caused by too many variables by regularization, variable selection is still meaningful.

Description: A brief introduction to a new SOA research on liability-driven investment benchmark model.

Description: The asset portfolios of retirees’ serve many purposes including current income, liquidity, purchasing power, estate, tax minimization, and so on. Multiple objectives with different levels of importance lead to a complex asset allocation problem for retirees. Existing asset allocation methods normally focus on a subset of the multiple objectives of retirees in an approximate way. The simulation-based multiple objective asset allocation method objectively assesses each allocation plan against multiple objectives in a consistent way and provides a more holistic picture of possible outcomes. This information is critical for finding the optimal allocation plan. The optimization is based on the weighted performance relative to multiple objectives. At the same time, it requires more inputs and advanced modeling.

Description: Economic risk analysis has two dimensions: time and frequency. Asset return varies by time because of economic cycles and economic structural changes. Risk is higher during economic recessions and smaller during economic expansions. In addition, different economic structures may exist at different time scales. Risk measures calculated based on daily, weekly, monthly and yearly historical data can be very different. The appropriate frequency depends on the time horizon of the analysis. Therefore, it is important to measure economic risk at both the time level and the frequency level. However, time series analysis and statistical analysis that have been widely adopted in economic risk analysis focus on only one of the two dimensions. In this paper, wavelet models are used to enhance both time‐invariant analysis and time‐variant analysis. Wavelet models can systematically analyze risk by time and frequency and provide richer information than time series models. Using wavelet models, risk measurement can easily be adjusted based on time horizon in a consistent way. Equity VaR estimation is selected to demonstrate the application of wavelet analysis in risk management.

Description: A brief introduction to a new SOA research on liability-driven investment benchmark model.

Description: Enterprise risk management (ERM) has grown fast in the 21st century, thankful due to a variety of extreme yet unfortunate risk events and regulatory changes. However, challenges continue with many other initiatives competing with ERM for resources within an organization. In this month’s newsletter, we revisit this old yet still important topic and bring a few relevant reports to your attention.

Description: This article discusses how unemployment and underemployment information can help improve economic forecasts, insurance assumption setting and risk management. It summarizes the current modeling practices of unemployment which shows a deficiency of using unemployment/ underemployment information in actuarial works. It also gives three examples of using unemployment/underemployment information for economic assumption setting, new business growth prediction and stress scenarios related to the labor market.

Description: Article describing the most recent Risk Management Section research updates

Description: This article introduces a non-traditional type of economic scenario generator that relies on dynamic stochastic general equilibrium (DSGE) models are complex macroeconomics models and have been increasingly popular in central banks for analyzing monetary policies.