Using Predictive Models and Risk Objectives for Portfolio Weights

Summary

The document addresses how to incorporate technical indicators, such as RSI, ATR, and moving averages, alongside historical returns when assigning portfolio weights. Its central suggestion is to use a model that converts selected inputs into expected returns for each asset, rather than assuming expected returns must equal their historical averages. Those forecasts can then become inputs to a portfolio optimization process, such as a Markowitz-style framework.

Risk need not be represented only by conventional variance: the response points to alternative risk measures and cites work on conditional value at risk objectives and constraints. The description remains conceptual and does not specify a forecasting model, training procedure, optimization constraints, or validation method. It therefore offers a framework for connecting predictors to portfolio construction, while leaving implementation choices and the reliability of forecasts to the researcher.

Key ideas

• A forecasting model can use technical indicators and historical data to estimate expected returns by asset.
• Portfolio optimization can use those forecasts instead of treating historical returns as expected returns.
• Optimization objectives can use risk measures beyond variance, including conditional value at risk.
• The proposed framework leaves model design, constraints, and forecast validation unspecified.

Tags

# Portfolio construction


# Portfolio construction












Suppose I have 3 stocks. Their historical returns and some variables like RSI, ATR, EMAs for all 3 of them. The goal is to compute the weights each stock should have in a portfolio. If I do something as simple as markowitz portfolio, it would mean I am only using the historical returns of the 3 stocks and ignoring other factors like RSI, ATR, EMAs.

My question is , how do I incorporate the historical stock returns and other factors and come up with a model that predicts the weights of the stocks in the portfolio?

## Answer by jaamor (score 3)

https://quant.stackexchange.com/a/22455

Markowitz portfolio optimization and variations of it usually boil down to maximizing expected returns while constraining the aggregate of your preferred measure for the portfolio.

The expected return does not have to be equal to the historical return in your model. You can feed in the factors of your choice within a model that outputs the expected return for each asset.

In literature, it is usually an alternative measure of variance that is used to feed into portfolio optimization.

This publicly available paper has a good overview of developments in this area in its introduction, if you want to read more: Porfolio Optimization with conditional VaR objective and constraints by Pavlo Krokhmal, Jonas Palmquist, and Stanislav Uryasev.

Shown in full with attribution under the source's licence. Licence: CC BY-SA 4.0 (Stack Exchange)

This summary was written by Stratmill's research agent from the original; it is not a copy of the source.