Staying out of declining markets

General
21

The original Shiller CAPE methodology compares the CAPE relative to its long term average of 154 years ignoring structural changes in markets over time. The CAPE-MA35 improves upon traditional CAPE analysis by normalizing CAPE relative to its own 35-year moving average. This adjustment allows the ratio to adapt to evolving macroeconomic conditions-such as changing interest rate regimes, globalization effects, and technological shifts-thereby offering a valuation metric that better reflects current structural dynamics rather than a fixed historical mean. 35 years was chosen because it includes at least 3 business cycles.

Importantly, the initial model specification and hypothesis formation were conducted using data from 2000 to 2025. The framework was subsequently evaluated on an independent historical period spanning 1928 to 2000, without modifying parameters or decision rules. The consistency of results across these two distinct samples provides out-of-sample support for the original hypothesis and further reduces the likelihood that the findings are driven by sample-specific effects or data-mining bias.

I have a designer model in waiting using the methodology which you can see early February 2026. Ultra CAPE-MA35 ETF Rotation Strategy.

1 Like
22

I understand that market timing is a loaded topic and a double-edged sword not to be used lightly. In my case, it’s just one of several defensive layers and is applied to only a small portion of my holdings.

It’s mostly canary-style logic, progressive; not binary, and I don’t see it failing spectacularly. If global markets begin to deteriorate meaningfully, I trust it will nudge me to be a bit less microcap-centric—and I expect I’ll be glad it did. I fully accept that it may occasionally switch on or off late, or imperfectly, but I still trust it in aggregate.

Between the 3×-bear-paired microcaps (constant protection), my SPX portfolio-level hedge (designed to catch fast crashes at a modest cost), this canary-timing sleeve (meant to shift subtly toward safety during systemic, drawn-out stress), and my ETF sleeve (which is negatively correlated to much of the rest), the combined effect is that I can run the portfolio just a bit hotter than I otherwise would.

Because this timing sleeve is built from a set of ‘rarely triggered’ hedge concepts, I view a negative outcome as only modestly worse than permanently pairing the microcap sleeve with cash.If it works well, it should slightly reduce CAGR while meaningfully reducing max drawdown. If it doesn’t work particularly well, it likely behaves like a simple 90/10 equity-cash pairing—which is something I arguably should be doing anyway. If it fails spectacularly, I’ll wish I had simply chosen the 90/10 cash pairing (and hope my other sleeves can save the day), but I see that outcome as least likely. Taken together, it feels like a bet worth making.

More broadly, I think some of what we know works at the individual-security level can also be applied at the portfolio level.

For example, Yuval, even if a momentum filter were applied to your models, isn’t it likely we’d happily have landed in your Euro model recently over others—much in the same way one would prefer holding a 99-ranked stock over a lesser-ranked one?

Or am I oversimplifying the analogy and overestimating how well this carries over to the portfolio level?

23

When the markets decline, I often wonder how many investors stay with their mechanical investing strategy, or begin to introduce other analysis… I know thats the case for me and it always ends up dragging down my returns.

3 Likes
24

While this is commendable, what would have happened if you'd looked only at 1928 to 2000, without considering what came later? Would you have come up with a different number of years?

Is a business cycle typically 11 years? I thought it was about four to seven years. Wouldn't 28 years have been sufficient? If you'd tested this in 2000, when business cycles had typically been much shorter, would you have come up with a different moving average period? What happens if you use 25 years instead of 35? Does it make a big difference or an insignificant one?

Why not corporate bonds? Aren't those what most people would have hedged with in the 20th century?

1 Like
25

It's quite small, under $10M AUM, I think.

1 Like
26

Why? I would think microcaps would do spectacularly well in the aftermath of a crash and wouldn't be as badly affected as overpriced megacaps during one. I suppose it depends on whether or not the crash is associated with a major tightening of credit, but there are plenty of microcaps that are not debt-burdened. An underpriced small ship builder or maker of handheld radios or provider of technical equipment to cinemas is likely to be more impervious to a market crash than an overpriced large company whose business model depends on massive leverage or selling shares while earnings growth tries to catch up to revenue growth. Or do I have that wrong?

Yes indeed, to some degree. It depends on whether you use the North Atlantic universe with a ranking system or not. If you use different ranking systems for Europe and North America, or for Canada and the US, then you won't get this effect. The only way it really works is to use the North Atlantic universe alone.

1 Like
27

Passive investing is just another name for passive speculation. Many large caps reach very large valuations, get more passive flows then collapse after reaching ridiculous levels like we just saw on some of the quantum names or probably palantir. Will be a stock by stock case as usual and a lot would depend on the reason behind the crash. A US centered recession typically would impact small caps more since more of their sales are in the US, but also forced liquidations and high valuations can wreck havoc on large caps. I would agree.

1 Like
28

The core idea is to use broad, protective momentum as a regime sensor, not a return-maximizing or predictive tool. When market-wide conditions are healthy, it stays out of the way and allows higher-volatility strategies to operate normally. When momentum across diversified “canary” assets deteriorates in a sustained way, it signals rising systemic risk and progressively nudges exposure lower. Precision isn’t required—late signals, partial whipsaws, and missed bottoms are acceptable—because the logic doesn’t need to be perfect to be helpful.

To reduce model-specific or overfitting risk, the hedge signal isn’t driven by a single rule set but by a small composite of related momentum and risk-off frameworks with slightly different logic and sensitivities. The goal isn’t to avoid every drawdown or flip risk on and off, but to modestly scale exposure during periods when risk-reward is structurally worse, smoothing outcomes and allowing the overall portfolio to run hotter than would otherwise be comfortable.

And if someone simply can’t get comfortable with the idea that canary-style signals can translate across asset classes, I completely understand—that skepticism is reasonable. This approach isn’t meant to be universal or convincing on faith; it’s just one component among several that I’ve found logical and complementary in aggregate. Finding Charles’s recent logic was particularly helpful in taking pressure off over-allocating to this sleeve, which—like any defensive layer—can be wrong at times.

29

I’ll admit it gives me pause to be somewhat on an island on this one—will need to take a closer look at some things.

30

Maybe doing more of an industry switch/tilt might make sense too, rather than market cap alone I do some of that when I want to be more defensive

31

Thanks, Santiago — I may not have been clear in my wording. I didn’t mean reducing microcap exposure in favor of large caps or other assets; all of my individual stock exposure is microcap.

My only other equity exposure is through a meaningful allocation to tactical ETF models, which manage risk independently and in which I have a fair amount of confidence. Within the microcap sleeve itself, one tranche is permanently invested but paired with TZA, a structure I’m reasonably comfortable with. A second tranche adjusts exposure up and down based on a set of canary indicators intended to reflect broad market conditions.

The fair criticism is whether those indicators are optimized or overfit — and I’m sure that, to some degree, they are. That said, the design is intentionally redundant but independent across layers, and exposure is rarely reduced much less taken fully to cash, which gives me some confidence in its robustness.

Put differently, if aggregate “risk-off” exposure averages around 10%, even applying that reduction randomly (with no signal at all) would likely still produce solid returns, just modestly muted. If the logic has any real regime sensitivity, drawdowns should improve at the cost of a small amount of return. It’s really only in the case of negative skill that this sleeve becomes potentially harmful.

I view this less as return-predictive market timing and more as risk-state modulation. I realize that calling it something fancier doesn’t reduce the odds of overfitting, but given how subtly and incrementally it adjusts exposure, I think it’s a more accurate description of what I’m trying to do.

32

Got it- thanks for explaining. I personally think it makes sense to act differently in different environments recognizing that we of course cannot always predict everything ahead of time and that the past is the past. Nature does the same. As Charles Darwin would say, “It is not the strongest of the species that survives, nor the most intelligent, but the one most adaptable to change”.

The way I look at investments these days is I need to think about who will buy it from me (if I intend on selling since sometimes I do not) and how easy it will be for them to spot what I spot. I have bought things in frontier markets before that remain undiscovered for a very long time for example

2 Likes
33

There is no data available for corporate bond funds for the full period from 1928 to 2000. I constructed a synthetic Treasury fund from the dividend yield shown for 10-yr Treasuries in the Shiller spread sheet. Gold is only of interest after 1973 when it could be traded freely and 1974 when US persons were allowed to own gold again after they were prohibited by executive order in 1933 from owning gold.

34

What difference does it make whether the out of sample period is before the in-sample period to validate a strategy?

35

Because of credit risk corporate bonds behave very differently to treasuries as well in terms of hedging

1 Like
36

What bond funds do you know of which were available during the Great Depression to hedge with, and historic data for them?

37

Try Ibbotsonn SBBI for US treasuries data -that is all you need. But basically you do not need a backtest to know corporate bonds have credit risk they crash with stocks while us treasuries have lower losses or even huge gains depending on the inflation rate. For example treasuries went up 20% in 2008 while corporates sank. One of the main reasons the quantum fund did well is they had both treasuries and stock positions concurrently via leverage. For a quick look at the yearly returns you could glance at nyu stern data online for free. My point is one is better off hedging with treasuries over corporates

1 Like
38

Correct, that is why I used 10-yr Treasuries for my backtest starting in 1928. Shiller provides monthly yields for this and one can construct a synthetic bond fond from this.

But Treasuries are not always the go-to for hedging. Just look at the performance of IEF from July 2020 to October 2023 - minus 23%.

39

Yes of course. Basically I agree with using them. For that time span it was because of inflation in that case. They do not hedge inflation. They hedge lower than expected growth and lower than expected inflation. Ideally you pair with assets with offsetting drivers. Assets that benefit from higher than expected growth and assets that benefit from higher than expected inflation.

1 Like