I looked into this about 10-12 years ago. To calculate Optimal f you need past trading history. If you have a system then you can also use historic trades, although I would certainly add a cushion as your biggest loss is always in the future. Lastly, I would personally trade more conservatively and choose a smaller bet size, but there is a lot of value in calculating the optimal size to bet to better understand position sizing.
Indeed. I like his idea of insuring your portfolio w options to guard against such black swan events Vince does mention ‘fractional f’ across some of his works (inflections points across his curve where acceleration of reward > acceleration of risk), and I may study that tradeoff further But currently I prefer his approach of using subaccounts w periodic rebalancing. If you run a portfolio of various strategies, it seems like the perfect way to guard against ruin. And it’s more intuitive to me
Yes - non-correlated strategies with their own allocation with periodic rebalancing makes a lot of sense. It's been so long since I read it (and I don't practise it now), but I seem to recall there was also a portfolio-level approach. Was that in his second book?
Same book, but I haven't reviewed that chapter yet. I'll respond once I do. What did you dislike about his approach, and what do you instead?
No dislikes with the optimal f approach, it's mathematically correct It is however applicable for systematic trading where you are able to calculate the inputs, and my trading has been on and off over the years and partly discretionary or through allocation so it's not been possible/relevant.
"f" probably stands for f...up. You ability to scale is based on your biggest "f". Your already using a rough version of it without being aware.
updating this thread. according to Sinclair, there's a Kelly variant designed for mean reverting spreads, as researched by boguslavsky & boguslavskaya (2004) and Liu & Longstaff (2004). its based off a simple Ornstein uhlenbeck MR model and assumes Gaussian distribution, but its a start. if one were so inclined, one could find Kelly for this model and simply convert to Vince's formula via the conversion factor. not sure if ill be doing it myself, but its still good info. haven't seen this idea mentioned anywhere else so hopefully someone somewhere finds this useful