MacropoiesisBETAGrand Tour

Macropoiesis is a research and learning platform. This lab is built on the thesis of Burton Malkiel's A Random Walk Down Wall Street: that short-horizon price changes carry almost no usable memory, and that the patterns we see in charts are mostly the pattern-hungry eye at work. Nothing here is investment advice.

Exhibit 1 — how the fakes are made

• A real round shows 60 consecutive trading days of an actual large-cap stock, normalized to start at 100, with dates and name withheld until you answer.
• A simulated round draws 59 daily log-returns from a normal distribution whose mean and standard deviation are measured from the same stock's five-year history — so volatility alone can't give the fake away. The coin-flip flavor instead steps up or down by one standard deviation at 50/50, Malkiel's original classroom construction.
• The answer is stored server-side until you guess; the seed of every round is revealed afterwards so any chart can be regenerated.
• After 20 rounds you get a two-sided binomial test of your accuracy against coin-flipping. Community accuracy across all players is shown next to yours — expect both to hover near 50%. That is the exhibit.

Exhibit 2 — the detectors

• The path is 250 steps of ±1 coin flips from a seeded generator running in your browser (the seed is displayed; the same seed always redraws the same chart).
• Detectors: 20/50-bar moving-average crosses; support/resistance as price levels touched three or more times within a 1.5% band; double tops as two peaks within 2% separated by a ≥3% trough; head & shoulders as three peaks with the middle one highest and the outer two within 3% of each other.
• These rules are deliberately the simple, honest versions of what chart lore describes. They fire constantly on coin flips — which is the point.

Exhibit 3 — the statistics

• Ljung-Box Q(10) asks whether the first ten autocorrelations of daily log returns are jointly zero.
• Wald-Wolfowitz runs test counts sign streaks: too few runs means trending, too many means mean reversion.
• Lo-MacKinlay variance ratio at 2, 5, 10 and 20 days compares multi-day to one-day return variance (≈1 under a random walk), with the heteroskedasticity-robust z* so volatility clustering alone doesn't reject.
• The six p-values are Holm-adjusted as one family, and the criteria above are fixed in code before any data is seen. A rejection means statistical predictability — which historically has rarely survived trading costs; that distinction is printed on the verdict itself.

Exhibit 4 — the monkey cohort

• Twenty seeded random 15-stock portfolios from a fixed ~100-name large-cap board, equal weight, one-way commissions charged at entry and on annual-rebalance turnover, simulated over five years against SPY through the same engine. The seeds are fixed, so the cohort only changes when prices do.
• The persistence matrix ranks the monkeys each calendar year and tabulates where each quintile lands the following year. Randomness predicts uniform rows and a near-zero rank correlation — the same test the fund-persistence literature applies to managers, with the same result.
• Survivorship caveat, stated honestly: today's board holds today's large caps, so the level of monkey returns is flattered; the dispersion and the non-persistence — the two lessons — are unaffected.

Exhibit 5 — the audit

• Every signal old enough for the longest horizon is compared, close-to-close, against the same stock's average forward return over the period — a dart thrown at the same board. SELL signals are scored on the inverted return.
• Significance comes from a symbol-clustered bootstrap (symbols resampled first, then entries), because five overlapping signals on one stock are closer to one observation than five.
• The six (side × horizon) p-values are Holm-adjusted as one family, and a cell only counts once it holds enough mature events — so the verdict starts INCONCLUSIVE and earns its way to anything stronger. Both refresh weekly.

Exhibit 6 — how the Mirror replays your purchases

• Each holding's purchase date and cash amount (quantity × purchase price, or the close on the purchase date when no price was recorded) buys SPY at that date's close instead. Both sides are then marked close-to-close to today. Lots without price coverage, or older than 10 years, are skipped and the count disclosed.
• This is deliberately backward-looking only — no cost basis feeds any forward-looking feature on this platform, and the Mirror never leaves this page. Non-USD lots are included without currency conversion and disclosed; treat their rows as approximate.
• The result is cached for a day and recomputes automatically when any holding changes.

Exhibit 7 — the arithmetic of costs

• Net return = gross return − expense ratio − advisory fee − turnover × an assumed 0.30% cost per 100% of annual turnover (spreads + market impact + realized-tax drag proxy). Both paths compound the same $10,000; the headline is the share of terminal wealth consumed.
• The glidepath is the classic illustrative age rule (equity ≈ 110 − age, split against bonds/cash/real assets), with long-run assumptions shown in the panel — stocks 7%/16%, bonds 3.5%/6%, cash 2%/1%, real assets 5%/12%. Portfolio volatility is a weighted upper bound (correlations ignored — a documented simplification). Illustration, not advice.

Exhibit 8 — the museum's charts

• No API serves tulip prices: each episode's series is a hand-authored, stylized shape normalized to peak = 100, embedded as static data. Peak-to-trough magnitudes, dates and recovery times follow the standard historical accounts; the points between them are illustrative.

What to learn from it

• Play twenty rounds of Exhibit 1 and read your binomial p-value — the feeling of "this one looks real" meeting the arithmetic of it.
• Reflip Exhibit 2 five times and count how many "textbook" formations the coin produced.
• Run Exhibit 3 on your favorite stock and on an index ETF — most days, both fail to reject; when one does reject, ask what round-trip costs would do to the edge.