Can Machine Learning Help Pension Funds?

The introduction of machine learning models has given institutional investors a way to manage their portfolios that is better adapted to their needs

Can Machine Learning Help Pension Funds?

Since 1997, a pension system with defined contributions has been operating in Mexico. Through individual accounts, the resources from the contributions of employers, employees and the government are invested in various financial instruments, both in the Mexican and international markets, through pension funds called SIEFOREs.

Promoted by the World Bank in the mid-1990s, the introduction of the defined-contribution pension scheme produced an increase in the number of institutional investors in the financial markets of the diverse countries that implemented the new system.

Today, pension funds are believed to play an important and positive role in a countries’ capital markets and economy. Moreover, several studies have been carried out to identify whether there are differences in the impact of pension funds in developed and emerging countries.

Improving retirement conditions

The reforms to the pension systems implemented in Mexico since 1992, and in particular the startup of the Retirement Savings System (SAR), have resulted in an improvement of the country’s financial sustainability. However, the conditions in which workers retire must be improved. To this end, some reforms to the system must be implemented to adjust retirement conditions to the nation’s labor and demographic dynamics.

According to the Inter-American Development Bank (IDB)[1], reform proposals should be organized around three core strategies:

1. Improving system governance

2. Increasing the coverage and level of present and future pensions

3. Reducing imbalances in the transition between systems

Driving system savings with artificial intelligence

Within these three core strategies, there are additional proposals to drive savings in the system, such as reducing operating costs, establishing an automatic voluntary savings mechanism, and improving savings efficiency by improving the investment structure.

In this regard, it becomes increasingly important for pension funds to have additional tools and models to optimize their portfolios according to their objectives and needs. For this type of institutional investor, a correct long-term portfolio management should be the most important focus.

With the introduction of machine learning models, long-term portfolio management has changed. Now, large institutional investors have found a way to manage their portfolios that seems to be more adapted to their needs, compared to traditional models based on financial theories.

Many such institutional investors have already adopted artificial intelligence for their portfolio management and diversification decisions.

Additionally, in the case of pension funds, there are studies that, supported by machine learning techniques, not only seek to solve portfolio management optimization problems, but also attempt to forecast the behavior of accumulated resources and thereby estimate the date on which each contributing employee can decide to retire from working life.

For example, in addition to data analysis technologies, an automatic learning tool called boosted decision tree -due to the proximity of this technique to that applied in financial forecasting- has been used to estimate the resources accumulated in the pension fund of a given affiliate according to its own economic level and growth aspirations. The model was even validated with a pension fund management system in Peru, and the results were a very good approximation to reality.

Some fund operators have also shown an interest in machine learning to model the early withdrawal of resources at a given time, thereby enabling adequate planning. Algorithms such as vector machine, logistic regression, and random forest have been used with data from private pension plans to predict whether a person will be able to retire before or after the age of 65, based on their individual characteristics and macroeconomic factors.

Finally, due to the demographic conditions in Mexico, where the economically active population will continue to increase progressively, SAR assets have grown rapidly and are expected to carry on doing so. Therefore, analyzing the impact that the growth of such assets will have on Mexican equity markets and debt instruments is paramount in order to visualize the improvements required by the current investment structure.

*The contents of this article are based on a chapter we authored on pension funds in emerging markets, included in the book Data Analytics Applications in Emerging Markets (Springer 2022).

The authors are professors at EGADE Business School (Martha León Alvarado) and at Universidad Autónoma Metropolitana (Beatriz Mota Aragón).

Article originally published in Alto Nivel.

[1] Diagnóstico del sistema de pensiones mexicano y opciones para reformarlo (2019)

Go to research
in your inbox