In chapter 2, Darst tries to carve up the space of approaches to 'asset allocation' through dimensions of style, then how strategic the approach is, and finally how quantitative the approach is. As I mentioned in the last post, I think the 'style' dimension is bogus. This in the limit can be replaced by owning the market of available strategies in toto), in their market weights, and then by implementing risk appetite purely through levering the in toto portfolio. Next his seemingly clear quantitative versus qualitative distinction breaks down too - for an ideal strategy allocation algorithm, the parameterisations are empirically calibrated and the discovery of new strategies are qualitative, whereas ideally the implementation, given a broad parameter set, ought to be quite algorithmic and computationally tractable. Again ideally, the re-allocation decision might in theory be near-real time.
Finally, the dimension of 'strategic' v 'tactical' is the difference between Kant and Machiavelli.
I think you want the algorithm to be as autonomous as possible, and to make a call on the strategic/tactical dimension based on the following inputs: where you are on your own expected wealth process and your expected lifespan. Your spend process ought to follow from these two, and shouldn't count as an input. Likewise this set of input parameters can be used in the determination of how much leverage to use (how long do we think it will take us to get there). Your expected spend (and the lumpiness thereof) is really a (time-dependent) constraint on the volatility you desire on your wealth process.
The starting point (the long term equilibrium point) would be based on the maximum likelihood weightings, based on as much data as there is available for the strategies. If one then subsequently had a model of strategy cycles, then that would be burned in too, to a degree proportional to one's confidence in the cycles model. The mean value theorem guarantees that your long term equilibrium parameters are a good starting point, in the face of no certainty about cycles at all.
Crescet, titubit and fugit are facts about you. Curcuit and the long term equilibrium weightings are parameters of the strategies.
The starting point (the long term equilibrium point) would be based on the maximum likelihood weightings, based on as much data as there is available for the strategies. If one then subsequently had a model of strategy cycles, then that would be burned in too, to a degree proportional to one's confidence in the cycles model. The mean value theorem guarantees that your long term equilibrium parameters are a good starting point, in the face of no certainty about cycles at all.
Crescet, titubit and fugit are facts about you. Curcuit and the long term equilibrium weightings are parameters of the strategies.
No comments:
Post a Comment