1: \begin{abstract}

2: We present a parsimonious neural network approach, which does not

3: rely on dynamic programming techniques, to solve dynamic portfolio

4: optimization problems subject to multiple investment constraints.

5: The number of parameters of the (potentially deep) neural network

6: remains independent of the number of portfolio rebalancing events, and in contrast to, for example, reinforcement learning, the approach

7: avoids the computation of high-dimensional conditional expectations.

8: As a result, the approach remains practical even when considering

9: large numbers of underlying assets, long investment time horizons

10: or very frequent rebalancing events. We prove convergence of the numerical

11: solution to the theoretical optimal solution of a large class of problems

12: under fairly general conditions, and present ground truth analyses

13: for a number of popular formulations, including mean-variance and

14: mean-conditional value-at-risk problems. We also show that it is feasible

15: to solve Sortino ratio-inspired objectives (penalizing only the variance

16: of wealth outcomes below the mean) in dynamic trading settings with

17: the proposed approach. Using numerical experiments, we demonstrate

18: that if the investment objective functional is separable in the sense

19: of dynamic programming, the correct time-consistent optimal investment

20: strategy is recovered, otherwise we obtain the correct pre-commitment

21: (time-inconsistent) investment strategy. The proposed approach remains

22: agnostic as to the underlying data generating assumptions, and results

23: are illustrated using (i) parametric models for underlying asset returns,

24: (ii) stationary block bootstrap resampling of empirical returns, and

25: (iii) generative adversarial network (GAN)-generated synthetic asset

26: returns. 

27: 

28: \medskip{}

29: 

30: \noindent \textbf{Keywords}: Asset allocation, portfolio optimization,

31: neural network, dynamic programming\medskip{}

32: 

33: \noindent \textbf{JEL classification}: G11, C61

34: \end{abstract}

35: