MarkowitzModel strage behaviour

[dev-trendrating]

Hi, I'm trying to reproduce Markowitz optimization in java and I'm using MarkowitzModel class. I'm starting with simple inputs of 4 assets with ten yearly returns. I've generated the covariance matrix and expected returns to use as inputs. I've added a constraint min zero and max 1 on the four assets and I've specified a risk aversion of 1.
What sounds strange to me is that the optimal portfolio has always the form [0,0,1,0], the weight is always totally assigned to the asses with the greatest expected return.
I've the same algorithm implemented in Python using panda, numpy and scipy , same optimization function, same constraints and same risk aversion, and the result is similar to this [0.04313374816540931,0.45686630969477304, 0.25000001501466346,0.2499999271251542]
I can't say which one is correct, what is strange is that the form of the result of ojalgo MarkowitzModel is always the same and even changing the input it assign 1 according to the asset with the greater expected return.
Thanks
Filippo

[apete]

Maybe risk aversion means different things in ojAlgo and panda/numpy/scipy...

If you try different values (different order of magnitude) with ojAlgo, don't you get different results at some point?

[dev-trendrating]

scipy SLSQP/minimize function takes the following function to minimize (risk_aversion * portfolio_risk - portfolio_return) where portfolio risk is (wt)C(w) and portfolio return is (wt)(r) which look pretty much identical to the minimization function used by ojAlgo that's why my expectation was to have a similar result.
I've changed risk aversion in ojAlgo Markowitz model from 1 to 20 and now assign 1 to the asses with lower expected return