Trend Following Can Enhance Withdrawal Rates - Part 4

This is hopefully the last of my series on trend following and withdrawal rates covered in part 1, part2, and part 3. Not only the last, I am also going chartless here!  Note also that I am not really going to be explicitly talking about trend following at all (see part 3). But since I was on a roll with "perfect withdrawal rate" analysis, I thought I'd round things out with a nod to all those things that can suck the life out of a withdrawal program -- especially, in this post's case, an explicit "regime" of low returns superimposed on top of otherwise normal randomness.  None of this will be very science-y or systematic; I just wanted to see what happens for one set of assumptions with simple variations of that theme.

The Basic Plan

The idea: start with a (linear) return distribution that is vaguely similar to a 60/40 portfolio, then ding it with a proxy for fees and taxes, then ding it again for superannuation risk by adding some extra duration, then, finally layer on an explicit "extra" suppression of returns for x years, and finally bring in vol a little bit (could be from trend following, could be something else) to see how much of the withdrawal program can be redeemed after beating it over the head[1].  During all of that, calculate "perfect withdrawal rates" (PWR - the withdrawal rate that would have worked with hindsight to spend a dollar of wealth to zero in a world of random returns) along the way. Then look at the 5th percentile (i.e., the place where 95% of the path-dependent withdrawal rates were better) of the PWR distributions to see: a) how much the "dings" hurts in total at that percentile, b) how much each step contributes to the total, and c) how much vol reduction can help at the end.

1. Returns - start with m = .08 and s = .10 (arbitrary but not far off from VSMGX).

2. Taxes and fees - There are better ways but here I'll shave .015 off of returns. This is incredibly arbitrary and not very sophisticated but what the heck.

3. Superannuation - Add 10 years to the baseline (30 years -> 40)

4.  Regime - There may be better ways to do this but here I'll suppress returns inside the "sim" by an extra 2% for 10  years. In effect I think this is a type of double counting.  The sim by it's nature will have paths that have a string of bad returns at the beginning (that's ok) to which we will now add another layer of bad returns on top (i.e., the double-count).  That, I think, is not all that analytically supportable but then again those are the rules of this game so we are going to roll with it for now.

5.  Vol reduction - at the end I'll cut the standard deviation estimate by a percent 

The Results

		PWR
	Step	5th %tile	total chg	contribution
1	baseline	0.0540
2	taxes/fees	0.0450		45%
3	superannuation	0.0397		27%
4	regime suppress	0.0340	-37%	29%
5	vol reduction	0.0367	+8%

I thought the regime suppression would hurt more than it did.  The tax and fee hit, which I purposely made big and blunt, has an outsize contribution in this game/scenario…but I guess, in the end, that does not surprise me at all.  The 8% add-back by reducing std dev by 1pt is nothing to sneeze at but I guess I'd start with taxes and fees first. 

-----

[1] by any other name this would be Monte Carlo simulation. Or, rather, it is maybe better thought of as "inside-out" MC simulation where rather than a distribution of final wealth for a given spend plan we have a distribution of withdrawal rates for end wealth = 0.