CraigTester wrote: ↑Tue Jun 11, 2024 11:24 am
Kevin M wrote: ↑Tue Jun 11, 2024 9:26 am
HootingSloth wrote: ↑Tue Jun 11, 2024 8:42 am
CraigTester wrote: ↑Tue Jun 11, 2024 8:27 am
Worst case stock returns improve with duration, so as you say, where do you draw the line...?

At 30 years, worst case is 3%; at 1 year worst case is a 50%-type crash.

So far I've limited my duration to 20 years, as this is the inflection point where worst case = break-even.

So what is the minimum TIPS YTM that would motivate us to extend to 21 years, 25, 30, 35.....?

This really is getting to the question of the thread...

Are you familiar with the work of Pastor and Stambaugh in

Are Stocks Really Less Volatile in the Long Run?
This paper looks very carefully at the question of whether the range of investment outcomes over long time horizons is more or less uncertain than the range over shorter horizons, based on existing data

It concludes that stocks are substantially more volatile over long horizons, from an investor's perspective. There is a reduction in uncertainty over long horizons that is attributable to "mean reversion" of returns. However, this reduction is more than offset by the growing uncertainty over long horizons that results from unknown basic parameters (e.g., uncertainty about future expected returns).

In sum, if you pay close attention to the data, you would see that the answer to your question (at what yield should I invest in TIPS at various time horizons?) should actually decline as the length of the horizon increases. Of course, the exact numbers for break even should differ from investor to investor because their circumstances and goals differ. But of 20 year TIPS are good for you now, than 30 year TIPS should be even better if you pay close enough attention.

Already linked to this paper earlier in the thread, and recommended that everyone who belives what Craig believes should read it.

Let me lay out several premises, and see if we can find the point of disconnect, before we dive into all the greek letters.

**Premises**
1) Bogle's "Equation" is valid: Stock returns = Starting Dividend Yield + Earnings Growth rate + Percentage change (annualized) in the P/E multiple.

2) Per Shiller's online data, while certainly lumpy, on average, (Earnings + Dividends) have consistently grown over the last 100+ years.

3) P/E multiples have had wild swings over the last 100+ years, but "ultimately" net to a zero long-term contribution to returns.

Let me stop here, for a moment... Do you agree with everything so far...?

I don't really agree with premise (3) personally, but these kinds of premises, although intentionally vaguely stated, generally are consistent with the analysis in Pastor and Stambaugh (although their framework is both much more general and much more mathematically precise). I will try to translate what they are saying into this context, in case it is of interest to anyone.

One simple model of stock market returns would be to say that each year of stock market returns just pulls from an identical random distribution of returns with a fixed expected value and fixed variance. Under this kind of simple model, the variance over all time horizons is constant, and so you generally would expect an investor to demand the same TIPS yield for different durations when comparing TIPS to stocks over different horizons, all else equal.

When we say something like P/E multiples both have wild swings and yet "ultimately" net to a zero in the long run, we are talking about a kind of "mean reversion" in stock returns. (Note that over the last 153 years in the U.S., P/E multiple expansion has contributed about 3.5x to the value of the market, or about 0.8% per year annualized, so we haven't really reached the long run in this sense between 1871 and today). Instead of taking the view that the expected return at any time is a constant and that the distribution of future returns are identically distributed, as in the simpler model above, we believe that whenever returns in year t are higher than expected, expected returns in subsequent years are diminished in order to offset this positive "shock" and return to some sort of ultra long-term point of stability in valuations. Similarly, if there is a negative "shock" of below expected returns in year t, then expected returns in future years will increase to eventually offset the shock.

If you build a toy model of this kind and, crucially, assume that there are fixed and knowable parameters driving the mean reversion, e.g. assume there is a fixed and knowable "true" expected return for stocks in 2024, "true" value towards which valuations are supposed to revert, knowable time scale over which this reversion occurs, etc.), then you can show that the "true" long-term variance of stock returns does indeed decline over longer time horizons. In fact, you will predict (consistent with real world data), that this decline in variance over longer time horizons will show up in historical backwards-looking stock market data.

However, Pastor and Stambaugh point out that an investor making a decision in 2024 is not on the same footing as an investor who is looking backwards at whatever actually ends up happening by, say, 2054. The investor in 2024 only has the current data available in 2024 and not the true values of any of the parameters described above or any of the actual valuation ratios (and hence, in this kind of model, actual expected returns) in each of the years between 2024 and 2053. So, the question for the investor making a decision in 2024 is not what the "true" long-term variance is in 10-, 20-, or 30-year returns (based on the true values of all of the relevant statistical parameters), but what is the "predictive" variance based on the investor's knowledge in 2024 (based on the data available today to estimate those parameters).

In particular, the fact that expected returns are "mean reverting" means that there is a degree of uncertainty about what future expected returns will be for any given future year because the expected return of stocks in, say, 2050, depends on what valuations end up doing between now and 2050. Pastor and Stambaugh show that, under a relatively broad set of assumptions, the positive contributions to long-term "predictive" variance from this uncertainty about future expected returns (and uncertainty about other "true" values of parameters) more than offsets the negative contribution to long-term variance derived from mean reversion. So an investor sitting here, today in 2024, and only able to look at information that is observable in 2024, actually has a better idea of what the CAGR of the stock market between 2024 and, say, 2034 will be than what the CAGR between 2024 and 2054 will be. This is still true despite the fact that, under this mean-reverting model, when an investor in 2124 looks backwards they will observe higher actual variance in 10-year returns than in 30-year returns.

All of this means that, if you believe in this kind of mean reverting model, an investor today that is deciding whether to buy TIPS maturing 2/15/54 vs. 2/15/44 will eliminate more uncertainty by buying the 30-year TIPS than the 20-year TIPS and so should be willing to accept a greater opportunity cost (and a lower yield) for the 30-year TIPS than the 20-year TIPS, all else equal.

Building TIPS ladder for all residual needs and some wants after SS, pension, and paid-off house. Other wants from 5% constant percentage from Risk Portfolio (80/20 AA w/ 80% global + 20% US-tilt)