So this model just takes away the fact that we dont need to input the expected returns for each assets as required by efficient frontier based models… Also we should input the view on expected return (say bullish or bearish) and our confidence level on our view into the model…

The model comes up with a asset weightage different from the market and our allocation is quite diversified.

Could anyone please confirm if my understanding is correct? Also, Is short selling the only difference between the two models?

The starting point is the assumption that the benchmark portfolio is the optimal portfolio i.e can’t be beaten.

If the investor has no opinion, he should just take the benchmark.

If he disagree with the benchmark, BL shows a way to incorporate his views/confidence level into the implied market allocation via a reverse engineer process to arrive at a better allocation for himself.

You can check Financial Modeling by Simon Benninga for an Excel implementation of the process if it helps.

A Markowitz Mean - Variance optimization process is highly sensitive to the expected returns input. B-L tries to manage this sensitivity down , by using the weights in a global portfolio of assets ( and asset covariances) to guide in deciding the expected returns ( by reverse engineering ) .

Now that you have more realistic expected returns ( in a market sense ) , it is possible to modify the return expectations of some of the asset returns ( and only where the investor has a view or expertise) , then do a constrained MVO to get realistic weights according to a particular strategy like Long-Only .

Since you started with a world benchmark of assets you are likely to be pretty diversified in the final choices , and the expense of estimation is far less than you would have spent otherwise