Elephant Index · Open audit
The whole recipe is public — methodology, parameters, code, and every past position. Reproduce the Index bit-for-bit from public FRED data, check it was never repainted, then attack it. No black box, nothing held back.
Step 1 of 5 · The Tide
Central banks are mandated actors — they inject or withdraw liquidity based on system need, not price levels or investor positioning. That liquidity flows through credit markets into every risk-sensitive asset, the same way gravity acts on mass.
Sentiment shifts. Narratives fade. Liquidity and credit conditions move markets.
The numbers below are not a strategy claim — they are what 60 years of data shows happens to market returns when the tide is in versus out.
+17.3% / yr
+10.3% / yr
+10.3% / yr
+2.5% / yr
+7.2% / yr
+1.7% / yr
Avg annual return by regime · hypothetical / backtested · not a trading strategy
Step 2 of 5 · The Windows
The Index has two sensors: a cause sensor that reads liquidity — net liquidity and the yield curve — and an effect sensor that reads price, through the 10-month trend. Price is the effect, not a reading of liquidity: contemporaneous correlation between liquidity and price is near zero or slightly negative, because the Fed eases into weakness.
We don't trade a liquidity forecast — nothing here claims liquidity predicts returns. The chart below is descriptive: it shows how liquidity and forward returns line up across the record. The Index's only use of it: when liquidity agrees with a price break, the cause sensor acts on a shorter confirmation. Agreement buys speed. Nothing else.
Pearson r · liquidity growth vs. each asset's forward return · by months of lead time. Descriptive lead–lag structure only — the Index never positions on a liquidity forecast; liquidity shortens confirmation when it agrees with a price break.
Step 3 of 5 · The Numbers
Against buy-and-hold, the Index keeps the return and roughly halves the worst drawdown on the Nasdaq — same public data, better risk. Switch to Bitcoin below: it cuts the drawdown about a third (−75% → −50%), not half — and still beats hold.
CAGR
…
…
Sharpe ratio
…
…
Max drawdown
…
…
Calmar ratio
…
…
Time in market
…
…
Position changes
—
…
Hypothetical / backtested · not financial advice · data through 2023-12. The −23% Nasdaq drawdown depends on the 2008 exit, which hung on a one-month window on one indicator leg — a slow-topping crisis could present a similarly narrow window and be missed. Between crises the Index ≈ buy-and-hold. (Series starts 2003 because net-liquidity data begins 2002-12.)
Step 4 of 5 · Verify It
A Monte Carlo permutation test: keep the Index's exact invested-and-aside spells, shuffle when they happen across 10,000 random timelines. How often does random timing match the real record?
Almost never (p < 0.1%). Run it yourself. Switching the asset also updates the metrics table above.
Monte Carlo permutation test — the real Index's drawdown vs. 10,000 random schedules with the same invested/aside statistics.
This is one of the attacks — run the full battery in Step 5 for the rest.
Step 5 of 5 · The Verdict
The code is inspectable. The test ran live in your browser — not a screenshot, not a claim. Nothing is held back: the whole recipe is public. You could compute the current index status yourself. The subscription is for not having to.
Reproduce every past position
Clone the repo, run it, rebuild every historical monthly regime bit-for-bit from public FRED data — not summary stats, the actual Index.
Check we never repaint
Each month's index status is hashed and timestamped the day it's published; the plaintext reveals later and matches the hash. No hindsight.
Run the attacks and the failure map
The full attack battery — including the tests it fails — plus where and why the Index breaks. All published.
Compare it to the free rule
The free 10-month trend rule and its delayed variant are in the repo. Ours ties them on return; we don't hide that.
12 self-scoring tests, including the ones the Index loses. Runs on your device via WebAssembly Python — nothing uploaded. First run downloads the runtime (~15MB), takes 1–2 minutes.
All figures hypothetical / backtested — not financial advice.
Challenges
Isn't this just a moving average?
Mostly — and we open-source that. A free 10-month trend rule (Faber, published 2007) gets you most of the drawdown reduction on a single asset. We publish it, and we don't beat it on return or on 64 years of drawdown — we proved that on ourselves.
What the liquidity overlay does is fail in different places than the trend rule. The trend rule rides the first months of a fast crash; the overlay can ride a bear while the Fed is easing. Those failure modes are anti-correlated, so we blend the two — never betting the fund on which failure the next decade brings. That's a hedge against the unknown, not an edge.
So the recipe isn't the product. What you pay for is running it correctly for a decade, the verification trail, and the execution and tax wrappers on the fund side — none of which a moving average gives you.
You cherry-picked the asset and period.
The Index was built on the Nasdaq-100. It then held out-of-sample on the S&P proxy (back to 1962), Bitcoin (from 2016), and high-yield credit — assets and eras it was never designed for. Cherry-picking cannot explain cross-era, cross-asset consistency.
The timing is just luck.
The Monte Carlo permutation test above quantifies this directly. 10,000 random timelines with the same invested/aside spell counts rarely reproduce the real drawdown reduction — per-asset p from 0.7% to 7%, Fisher-combined p = 0.0022. That's not luck.
You fished thousands of parameter combos.
Deflated Sharpe ratios apply a haircut proportional to the number of trials run. After the haircut, the Index clears the bar on Bitcoin (DSR 0.99 — sensitive to the lookback window used, disclosed).
It's propped up by one lucky crisis.
Drop any single crisis — 1974, 2008, 2020, 2022 — and the risk-adjusted result holds. The edge is not a function of any one event.
You found a pattern with no reason to exist.
It has a structural reason: the Federal Reserve is a mandated, price-insensitive actor. It must inject liquidity when the system requires it, regardless of where capital is positioned. That's the structural reason the edge should persist — it's also the argument the attack battery below tries hardest to break.
Where is it weak?
On calm, diversified large-cap portfolios the risk-adjusted edge is not statistically significant. This is a seatbelt, not alpha. It is designed for liquidity-sensitive assets — the Nasdaq-100, Bitcoin, high-yield credit — not as a general-purpose index.