Journal / Case study · 11 min read
We tried more than 40 indicators and signals and a dozen strategy ideas. Almost none helped.
A case study of the indicators and strategy ideas we tested on rules-based ETF portfolios, why most did not improve returns or drawdowns, and who this approach is for.
By Lowdraft. Published October 20, 2026.
When people first see a rules-based portfolio, they ask the same thing we did: if the rules use a few indicators and work, would more indicators work better? We spent months finding out. We added indicators to our best Keels, swapped them in for the ones we had, and ran each change through the same tests. This note lays out the yardsticks, what we tried, and why the answer was mostly no.
Who this is for
Lowdraft is a software tool, not an adviser, and it does not manage money. The people who get the most from it tend to look like this:
- Investors with roughly $50,000 or more who want a clear, written plan for their ETF portfolio.
- People who pay an adviser 1% or more a year and want to know what they are paying for. Run the same yardsticks on both and compare. Maybe your adviser comes out ahead. Maybe you learn what to ask.
- People who want a smoother ride than stocks give. ETF Keels aim for shallower drawdowns than the market. Stocks can grow faster, and they also gap down on one earnings report. This is an alternative style, not an argument against owning stocks.
- People who prefer rules to gut feel and want to see how a plan would have behaved before risking money.
It is not for people who want guaranteed results, who cannot tolerate any loss, or who need personal advice. No Keel removes market risk.
Why fees are worth comparing
A 1% yearly fee sounds small. Over 20 years it is not. Take $50,000 that earns 7% a year before fees:
| Yearly fee | Value after 20 years | Cost versus no fee |
|---|---|---|
| 0% | $193,484 | — |
| 0.5% | $176,182 | $17,302 |
| 1.0% | $160,357 | $33,127 |
| 1.5% | $145,888 | $47,596 |
Arithmetic illustration with an assumed 7% yearly return before fees. Not a forecast, and not a claim that any provider's results are better or worse.
A fee is worth paying if it buys a good result or good advice. The point is to be able to check. That needs yardsticks, so we start there.
Our yardsticks
We rank every Keel against the S&P 500 (SPY) on the same dates, in this order:
- UPI (Ulcer Performance Index). Return divided by the Ulcer Index, which measures how deep and how long the portfolio sat below its previous high. It rewards a smooth ride and fast recoveries. This comes first.
- Sortino. Return per unit of downside volatility. It matters, but it does not see how long a slump lasts, so it ranks below UPI.
- Maximum drawdown. The worst fall from a peak. It tells you the worst stretch you would have sat through.
- CAGR. The steady yearly growth rate. We watch it, but a high CAGR with deep drawdowns is not what we are after.
- Several windows, not one. Full history, the last three years, the last year, and calendar years against SPY. A Keel that wins only in one window is suspicious.
- Rising trends rank higher. A Keel getting better over time scores above one that is fading.
A change had to clear a bar to count: a clear gain in UPI over the full history, a gain in both the first 70% of the data (where we designed it) and the last 30% (which we held back), no deeper drawdown, and little loss of return. We also made it survive realistic costs, a one-day trading delay, and the weekday check described in our rebalancing note.
What we tried
Across roughly 90 overlays on nine top Keels in the first round, and more than 165 on seven in the second, plus every idea since, here is the short version:
| What we added | What happened |
|---|---|
| Moving-average trend filters (SMA, EMA, zero-lag EMA) on single holdings | These are the foundation of our Keels and they work. Piling more of them on top did not add value. |
| ADX, +DI/−DI, MACD and ATR as bear filters on the market | Hurt nearly everywhere. Only a gate on our slow Income Keel helped (UPI 2.9 to about 4). |
| Fifteen further measures: distance from highs, relative strength, momentum variants, Donchian channels, %B, money-flow and volume measures, credit and defensive ratios, correlation, market breadth | Of 165 variants, three strict passes, all on weak starting Keels. Zero improvement on our top Keels. |
| A multi-part trend indicator (MACD with Bollinger bands, a momentum score, Donchian trend, RSI filter, Hull average) | It exited about 5 days after a peak against 14 for a 200-day average, which sounds great. But faster exits meant more whipsaws, and overall results were no better. Dropped into a Keel as the crash gate, UPI fell from 8.6 to 4 to 5. |
| VWAP, anchored VWAP, RSI divergence, volume profile, squeeze momentum | None beat a plain 200-day average as a trend signal. Divergence signals fired about 1.5 times a year, too rarely to matter. |
| Choppiness index, Bollinger bandwidth, efficiency ratio, adaptive average | They describe the market correctly. A 200-day trend signal earned 4.2% a year when choppy against 14.0% when trending. But using that to switch the rules off did not improve results. |
| Williams Vix Fix, a "bottom finder" for getting back in sooner | No edge: 20 days after a signal, ETFs returned +0.9% against +1.2% on an ordinary day. Re-entering on it made 2020 worse. |
| Letting each holding choose its own best moving-average length from its recent past | Worse than a fixed 200 days (median UPI 0.4 against 0.6). It fits the past and trips on the future. |
| A machine-learning score (ridge regression) to rank boost candidates | Its rankings had a slightly negative correlation with what came next. There was no predictability to use. |
| Dual momentum rotation among broad ETFs | UPI 0.4 to 4.0 and drawdowns of −23% to −53%. No protection. |
| Other weighting schemes (equal risk, momentum tilt, rank weights, caps) | Same as or worse than weighting by calm. Shifting weight toward the boost sleeve in strong markets raised return by 0.5 to 2 points a year but lowered UPI by 1 to 3. |
| Volatility targeting | Lowered return and drawdown together. UPI flat or lower. |
| Market-breadth triggers, extra crash triggers, more or fewer holdings in the core | Flat or worse. Choosing only the best 5 of 8 core holdings cost about 2 points of UPI. |
Why more indicators did not help
- Most indicators are the same information. Trend, momentum, MACD, ADX and moving averages all describe "price has been going up." Adding the fifth version adds little.
- Early signals come with false alarms. We could make the Keel exit faster. We could not do it without exiting at the wrong time too often. The cost showed up as return and UPI.
- A few days decide a lot. Much of the benefit over SPY comes from stepping aside in the 2020 crash. Anything that touches those few weeks swings results, and it is hard to tell whether the change is a gain or luck.
- Every extra dial is another chance to fit noise. We saw this directly: when we searched for the best group of core holdings on the first half of the data, the ranking barely held up on the second half (correlation about 0.19).
- The weighting was already near a plateau. Weighting by calm (inverse volatility) did most of the work. The other schemes moved return and drawdown together, not UPI.
What did help
A short list, each small:
- Weighting the two main sleeves by their calm rather than a fixed 60/40: UPI 8.6 to 9.2 on our lead Keel, a gain that held across window lengths and all five weekdays.
- A better defensive mix: cash-like funds always, plus the best two of three alternatives chosen by recent UPI. About +0.1 to +0.3 UPI.
- Swapping one stock-index fund for gold in the core: UPI 7.2 to 8.0 in the core Keel we tested.
- Requiring a stronger market (4% over 20 days, not 3%) before turning the boost sleeve on: UPI up in all 13 Keels we tested, typically by 0.3 to 1.4.
- Weekly rebalancing, as covered in our rebalancing note.
None of these is a magic indicator. They are changes to how the pieces are combined, and each one survived a held-back test.
What this means for you
- Be suspicious of an indicator pitch. If someone shows a chart where a new signal turned an ordinary strategy into a smooth, high-return one, ask for the held-back results, the costs, and the test period.
- Simple rules you can explain tend to last. A trend filter, a calm-weighting rule and a crash switch do most of the work.
- Expect an honest "no." A good testing tool will often tell you a change does not help. That is information, and it saves you from a worse portfolio.
- Compare on the same yardsticks. If you have an adviser, ask for drawdown, UPI or Sortino, and the benchmark over the same dates, and put the numbers side by side with any rules-based alternative.
Open this note as its own page
Hypothetical results from our research, prepared with hindsight and partly fitted to past data. Tests assume a one-day delay and trading costs; real trading adds slippage and taxes. Nothing here is investment advice, an offer, or a statement that any adviser's results are better or worse than a Keel's. Past performance, real or simulated, does not predict future results. You can lose money.