Black-Litterman Weight Instability and Benchmark Underperformance

Summary

The document describes difficulties implementing Black-Litterman with analyst-consensus views. The author reports portfolio weights that become negative or exceed the full portfolio allocation, as well as returns below a market-weighted benchmark. In one dataset, the underlying assets performed better than the benchmark before they were incorporated into the model, while the resulting portfolio underperformed. The author also observes that small changes in the view matrix or expected-return views can materially change the outputs.

These observations frame practical questions about input construction, model sensitivity, and portfolio constraints, but the document provides no answer or diagnosis. It does not specify the equations, covariance estimates, view confidence assumptions, rebalancing rules, transaction costs, or evaluation period needed to identify the cause. The reported underperformance alone does not show that the model is faulty or that the views lack value. The account is useful as a troubleshooting case, with the central caution that implementation choices and portfolio constraints can strongly affect Black-Litterman results.

Key ideas

• The author reports negative and highly concentrated weights from a Black-Litterman implementation.
• The resulting portfolios underperformed a market-weighted benchmark in the described data.
• Small changes to the view matrix or return views reportedly changed the outputs substantially.
• The document does not provide a diagnosis or a method for correcting the implementation.
• Performance comparisons require details about inputs, constraints, costs, and evaluation design.

Tags

# Problems with Black-Litterman: negative portfolio weights, and very poor returns


# Problems with Black-Litterman: negative portfolio weights, and very poor returns












I am trying to implement the Black-Litterman model using own-defined views matrix (from consensus analysts). However, I have encountered the problems of negative portfolio weights in some periods, and some extreme weightings >1 (at most 200%). All of the results end up in underperformance relative to the benchmark (which is simply the market-weighted stocks without strategy, i.e. passive index). In one data set, the cumulative returns of the assets (before implemented in the model) generates superior returns relative to the benchmark, but when after implemented in the model, it underperforms.

Not sure what I am doing wrong, as I am following the instructions. At lest the BL main formula works as intended using examples from other data set or articles. Additionally, the model is also sensitive to changes in the inputs, such as the changes in the P-matrix and Q-matrix. Slightly changing the Q-matrix from negative to positive returns changes the results of the Black-Litterman model.

Could anyone guide me in the right direction?

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.