Reinforcement Learning for Finance

Chapter 8. Dynamic Asset Allocation

Professional gamblers, who have to have an advantage, speak of “money management.” This refers to the tricky and all-important issue of how to achieve the greatest profit from a favorable betting opportunity. You can be the world’s greatest poker player, backgammon player, or handicapper, but if you can’t manage your money, you’ll end up broke. The sad fact is, almost everyone who gambles goes broke in the long run.

Poundstone (2010)

The world economy has grown at a decent enough clip over the past two decades, at more than 3% a year. Yet it has been left in the dust by growth in wealth. Between 2000 and 2020 the total stock rose from $160trn, or four times global output, to $510trn, or six times output.

The Economist (2023)

The challenge of asset allocation is a major problem in the financial domain, underscored by the vast amounts of money that individuals and institutions must invest. It is also a problem that started the quantitative revolution in finance with the seminal work of Markowitz (1952) on “Portfolio Selection.” In this paper, Markowitz proposes a purely statistical approach for composing portfolios as compared to, say, the fundamental analysis of companies and their stocks.

While the early work in this area focuses on the static, or nonrepeated, problem of allocating funds across different assets, a more realistic way of approaching asset allocation is in its dynamic, or repeated, form. Like algorithmic trading and dynamic ...