Up one percent for the week. Still gaining solidly each week. Not as strongly the market itself, but the strategy isn’t sitting out the rally either.
A reader mentioned that I should chart the allocations of the strategy since I started publishing it every week. I agree. So here is a chart of the performance since June 7th 2019 with the asset allocations over that period.
A slight shift from bonds into stocks today. Still holding quite a bit of cash. On April 17th, 2020, the strategy rebalanced to:
23% SPY , 29% TLT , 13% GLD , 35% Cash
Thanks for your blog and for sharing interesting knowledge!
Your weight assets graphs are smooth and it’s nice, you don’t need to radically rebalance the portfolio. And your portfolio is always diversified. The algorithm does not put all the money in one asset.
I have multiple important questions:
1. Geometric Return = mu – risk aversion * standard deviation ^ 2 / 2
Do you use risk aversion constant? Is risk aversion equal to 1 in your case?
If risk aversion = 1 the algorithm tends to invest all money in one asset, example:
https://drive.google.com/open?id=1dlV1_gGn3ibXEa3IQAKWCqYSBTP7srO7
2. What rolling window do you use to calculate mean return and covariance matrix?
3. Do you use some kind of exponential moving average?
1.) No risk aversion. At least none in a formula like that.
2.) I don’t project returns with historical returns. For covariance see: https://breakingthemarket.com/convergence-time/
3.) No. I’m thinking of maybe using one though.
4.) Not extremely knowledgeable with L2 regularizatoin, but unless missed something about it, no.
Thanks so much.
4. Maybe you use L2 regularization?
Thanks for the response. Article about standard deviation and correlation is clear. I’m also trying to use exponential moving covariance.
What data do you use for prediction of returns? Please share at least the direction of thought.
Very helpful to see the visuals. I was actually looking back on posts and trying to reconstruct to see over time how/when the cash (and others) shifted. Thanks for all this. Tom