Another half-a** look at trend following and retirement portfolios

This has probably been done a thousand times before, and I'm not totally sure I can do what I am trying to do here, but I wanted to hold the data in my own hands to see what it looks like.  Here in this post I wanted to:

1) Get some real market data and see how non-normal it looks using monthly data and the imperfectness of yahoo finance data, and then

2) Look at the relationship between market (S&P, which isn't really "the market") and some trend following data to see if I could see the "convexity thing" and then

3) See if I had anything to say about this particular look-see...


1. The market return data and fat tails.

This is yahoo finance data for S&P and VSMGX (vanguard 60/40 MF), monthly data, 1994+ (earliest data for vsmgx) this is monthly simple arithmetic returns not ln(r2/r1) using yahoo adj close total return data. Yes I realize there are pros and cons to this data and approach. The green circle shows the fat tail. Some of this fat tail disappears at larger aggregation units (yearly or longer) but it makes a point.  Red is a simulated (10k iteration) normal distribution using the mean return and sd from the data. I guess we can confirm that there is a fat tail, but this is not news.

2) The relationship between the market and a trend-following approach.

I grabbed some AQR data on time series momentum and threw it up against the S&P over the period 1985+ which is the oldest data in the AQR dataset.  I have some reservations about my approach as well as some quirky assumptions [1] but we can maybe proceed anyway. Here is a scatter of S&P vs the time series momentum data:

3) Do I have anything to say about this particular quick and dirty look-see...

Yes.

a) if portfolio returns are not normal and have fat left tails, and if that non-normality impinges portfolio return goals (even before we get to spending), and then if that non-normality really really impinges on goals after we consider consumption (see last post), and

b) if alternative risk strategies (e.g., trend following or momentum in its various forms) do, in fact, have convex return structures that can mitigate left tail risk without crimping too much upside and are more efficient than an options strategy (not explored here), which it actually looks like they do and are, then

c) alt-risk strategies like trend-following look like they have something pretty powerful to say to a retirement decumulation program.

I'm only scratching a surface here as an amateur but this is a worthy path to follow...  I still think that the shape of spending plans has more impact than asset allocation, but not considering these kinds of issues related to tail risk management seem like a type of tax on laziness.

See also:

Clare, Seaton, Smith and Thomas (2017a), Can Sustainable Withdrawal rates be Enhanced by Trend Following?

Hoffstein, C No Pain, No Premium, NewFound, Feb 2019
---------------------------------------
[1] The data comes from the AQR data set for time series momentum  https://www.aqr.com/Insights/Datasets/Time-Series-Momentum-Factors-Monthly

Idk, this approach does not make me very comfortable. Here are some concerns:

 - not sure I have synced up the right time frames perfectly at a monthly level
 - the AQR data is excess returns and S&P is yahoo finance total ret adj close.  Not a great match
 - not totally sure I can do what I'm trying to do with chaining returns for 18 months but...
 - I did chain the geometric return over 18 months. That's pretty arbitrary but I have reasons.
 - Supposedly time series momentum and trend following are not exactly the same but they kinda are
 - I have not bothered to understand this well enough to really do this kind of thing yet, but..
 - I only look at 1985+ because that's the earliest the AQR data goes

fwiw, the monthly returns are geometrically chained for 18 months and then annualized.  18 months is sort of arbitrary but it's also about the sweet-spot for me in holding trend following positions in the past.