Can you predict financial markets from the weather? El Niño shifts rainfall across whole continents and moves what the world grows, so the idea is not absurd. The answer here is no, and finding that out properly required a way to catch yourself being wrong, which then went on to kill the best-performing result in the collection. Two ideas held up. Three that appeared to were withdrawn.
| Experiment | Return for risk | Ann. return | Max fall | Beats search | Outcome |
|---|---|---|---|---|---|
| Sell in May, seven ways | 0.58 | 5.4% | 36.1% | 25.6% | Validated |
| Volatility-managed equity | 0.53 | 5.4% | 35.2% | 14.8% | Validated |
| ENSO onset as a discrete event trade | 0.41 | 6.1% | 41.2% | 9.4% | Failed |
| Crop-price momentum traded through ag-input equities | 0.34 | 5.0% | 46.6% | 3.8% | Failed |
| 12m time-series momentum, equity indices + FX only | 0.33 | 3.9% | 50.3% | 0.7% | Failed |
| AUD/NZD sized by the causal SOI z-score | 0.31 | 2.0% | 29.3% | 4.4% | Failed |
| Time-series momentum across futures, FX and equity indices | 0.29 | 3.4% | 50.3% | 0.3% | Failed |
| Long grains outright when La Nina is established | 0.27 | 2.9% | 37.2% | 3.5% | Failed |
| Long softs outright when El Nino is established | 0.25 | 5.1% | 110.4% | 1.8% | Failed |
| IOD innovation sizing a grain-futures L/S basket | 0.14 | 3.1% | 70.0% | 0.4% | Failed |
The best result in the project, and it was not real.
Buy the crops that have historically done well in the coming month and sell those that have done badly. It looked like the best result in the project until the price data itself turned out to produce the pattern.
Cleared every criterion. Still loses to buy-and-hold in dollars.
Own seven of the world's stock markets from November to April, then sit in cash until October. Over fifty-five years this grew $1 into $19.70 while the worst fall was a third of the account — against $36.80 and a loss of nearly two-thirds for simply staying invested.
Halves the drawdown. The Sharpe gain is not the point.
Hold less of the S&P 500 when the market has recently been turbulent, more when it has been calm. Over fifty-three years this cut the worst loss from three-quarters of the account to just over a third — while ending with roughly a quarter of the money.
Survived correction, then failed on effective sample size.
AMO is the Atlantic Multidecadal Oscillation, a roughly 65-year swing in North Atlantic sea temperatures whose warm phases bring more hurricanes. This tests whether its 60-year cycle prices coastal catastrophe risk eras. Tested at one and three months; 2 of 2 cleared the usual bar alone, none once the size of the search was counted.
Composing two winners produced a copy of one of them.
Two rules had already worked on their own: sit out the northern summer and hold shares only from November to April, and separately, hold less whenever markets have recently been turbulent. This combines them. The result was a portfolio identical to the first rule alone — the testing framework keeps total exposure fixed, so it cannot see the second rule at all.
The climate thesis at its strongest — four of six criteria.
Wait for El Niño or La Niña to formally arrive, then buy the crops that the weather should affect, and hold for six months. The strongest climate result here — though almost all of the profit came from a handful of months.
Significantly negative momentum, which is itself the evidence.
Buy the crops that have risen most over the past year, sell those that have fallen most. It lost money consistently — which turned out to be a clue about the data rather than about crops.
The strongest signal in the study had no mechanism at all.
Gold, Microsoft, the Nikkei and long bonds tested against the Southern Oscillation Index: they fire at 8.3% against the real hypotheses' 8.8%.
The instrument reading zero. Everything else depends on it.
Not a strategy — a test of the testing. It runs identical machinery but decides which way to bet on each asset by coin flip. If this scored well, nothing else on this site could be believed.
Where textbook inference and an honest null disagree.
The textbook trade — El Niño droughts Australia, wheat rallies — tested across six grain instruments and four horizons. Nothing survives, and the significant results carry the wrong sign.
Buy the crops that have historically done well in the coming month and sell those that have done badly. It looked like the best result in the project until the price data itself turned out to produce the pattern.
Own seven of the world's stock markets from November to April, then sit in cash until October. Over fifty-five years this grew $1 into $19.70 while the worst fall was a third of the account — against $36.80 and a loss of nearly two-thirds for simply staying invested.
Two rules had already worked on their own: sit out the northern summer and hold shares only from November to April, and separately, hold less whenever markets have recently been turbulent. This combines them. The result was a portfolio identical to the first rule alone — the testing framework keeps total exposure fixed, so it cannot see the second rule at all.
Hold less of the S&P 500 when the market has recently been turbulent, more when it has been calm. Over fifty-three years this cut the worst loss from three-quarters of the account to just over a third — while ending with roughly a quarter of the money.
Wait for El Niño or La Niña to formally arrive, then buy the crops that the weather should affect, and hold for six months. The strongest climate result here — though almost all of the profit came from a handful of months.
When crop prices have been rising, buy the companies that sell tractors and fertiliser, on the theory that farm income reaches them later. Mildly positive, and well within what chance produces.
The same trend-following rule, restricted to stock markets and currencies where the price data can be trusted. Still positive, still short of the bar, and it lost to simply buying and holding.
Bet the Australian dollar against the New Zealand dollar according to the Southern Oscillation Index, since the same weather tends to be wet on one side of the Tasman and dry on the other. Small, and inseparable from luck.
Buy whatever has risen over the past year and sell whatever has fallen, across currencies, crops, metals and stock markets. It made money, but less than simply holding the same things.
Buy wheat, corn and soybeans whenever La Niña is established, and hold nothing otherwise. Nearly all of the twenty-year gain arrived in two episodes: 2010 and 2021.
Buy coffee, cocoa, sugar and cotton whenever El Niño takes hold, and hold nothing the rest of the time. Over twenty-five years it gained about 5% a year, but the swings were wide enough that the result cannot be told apart from luck.
Over 25 years it returned about 3% a year. The plausible range of outcomes still includes doing nothing at all.
Buy the farming companies that have climbed most over the past year and sell the laggards. Over twenty-five years it went essentially nowhere.
Rather than deciding in advance which way each market should respond to El Nino, let the strategy learn each one from history as it goes, using only what was known at the time.
Over 27 years it returned about -0% a year. The plausible range of outcomes still includes doing nothing at all.
The same learn-as-you-go rule as the previous experiment, but driven by the atmospheric pressure measure rather than the ocean temperature one. Two readings of the same climate system disagreed sharply, which is the finding: a real effect could not do that.
Buy the crops that have risen most over the past year, sell those that have fallen most. It lost money consistently — which turned out to be a clue about the data rather than about crops.
Follow crop price trends, but only while the climate is unusually disturbed. Filtering for extreme weather made the losses worse, not better.
The comparison portfolio for the volatility rule: buy the S&P 500 and never sell.
The comparison portfolio for the trend and seasonality strategies: own everything, all the time, with each market sized so it contributes a similar share of the risk. Beating this is the minimum bar.
The comparison portfolio for trend-following on assets whose price data can be trusted: own all of them, all the time.
The comparison portfolio for the seasonal strategy: the same seven stock markets, held continuously instead of only from November to April.
The comparison portfolio for the farming-company strategies: own the whole basket continuously, measured against each company's home market so it is a bet on the companies rather than on shares in general.
The comparison portfolio for the ENSO-onset trade: simply owning wheat, corn and soybeans all the time. A weather strategy has to beat holding the same crops through every season to be worth anything.
Not a strategy — a test of the testing. It runs identical machinery but decides which way to bet on each asset by coin flip. If this scored well, nothing else on this site could be believed.
Tested 20 ways. 2 looked convincing alone, none once the number of attempts was counted.
The textbook trade — El Niño droughts Australia, wheat rallies — tested across six grain instruments and four horizons. Nothing survives, and the significant results carry the wrong sign.
Tested 24 ways. 2 looked convincing alone, none once the number of attempts was counted.
Tested 9 ways. Nothing beat chance.
Tested 6 ways. 2 looked convincing alone, none once the number of attempts was counted.
Tested 24 ways. Nothing beat chance.
Tested 12 ways. Nothing beat chance.
Gold, Microsoft, the Nikkei and long bonds tested against the Southern Oscillation Index: they fire at 8.3% against the real hypotheses' 8.8%.
Tested 18 ways. Nothing beat chance.
AMO is the Atlantic Multidecadal Oscillation, a roughly 65-year swing in North Atlantic sea temperatures whose warm phases bring more hurricanes. This tests whether its 60-year cycle prices coastal catastrophe risk eras. Tested at one and three months; 2 of 2 cleared the usual bar alone, none once the size of the search was counted.
AMO is the Atlantic Multidecadal Oscillation, a roughly 65-year swing in North Atlantic sea temperatures whose warm phases bring more hurricanes. This tests whether its warm phase = active hurricane eras → persistent insurer discount. Tested at one and three months; nothing beat chance.
AO is the Arctic Oscillation, which governs how far cold air spills south out of the polar region. This tests whether its cold outbreaks → utilities over market, winters only. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
ONI is the Oceanic Nino Index, the official measure of El Nino strength; and SOI is the Southern Oscillation Index, the air-pressure difference between Tahiti and Darwin used to track El Nino. This tests whether climate-index *disagreement* (SOI vs ONI divergence) = atmosphere-ocean decoupling → forecast busts → ag vol regime. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
AO is the Arctic Oscillation, which governs how far cold air spills south out of the polar region. This tests whether cold-snap months (AO deeply negative) lift utility earnings visibility. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
SAM is the Southern Annular Mode, the north-south shift of the westerly winds circling Antarctica; and ONI is the Oceanic Nino Index, the official measure of El Nino strength. This tests whether its ensemble agreement (SOI+ONI+DMI+SAM all aligned) sizes conviction. Tested at one and three months; nothing beat chance.
ENSO is the El Nino-Southern Oscillation, the Pacific climate cycle that swings between El Nino and La Nina. This tests whether its regime persistence (not level) prices ag-input capex cycles. Tested at one and three months; nothing beat chance.
ENSO is the El Nino-Southern Oscillation, the Pacific climate cycle that swings between El Nino and La Nina. This tests whether its → Indonesian equity complex (palm, nickel, rice imports). Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
ENSO is the El Nino-Southern Oscillation, the Pacific climate cycle that swings between El Nino and La Nina. This tests whether halloween strength is ENSO-conditional (does La Niña amplify Nov-Apr?). Tested at one and three months; nothing beat chance.
HVAC is heating and cooling equipment. This tests whether heat-wave summers pull HVAC orders. Tested at one and three months; nothing beat chance.
SAD is seasonal affective disorder, the winter dip in mood tied to daylight. This tests whether hemispheric SAD arbitrage: long the hemisphere entering spring, 6m alternation. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
ENSO is the El Nino-Southern Oscillation, the Pacific climate cycle that swings between El Nino and La Nina. This tests whether market-implied ENSO: PCA of climate-exposed equity residuals leads official indices. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
NAO is the North Atlantic Oscillation, the pressure seesaw that steers Atlantic storms into or away from Europe. This tests whether its drives European winter heating demand. Tested at one and three months; nothing beat chance.
Tested at one and three months; 2 of 2 cleared the usual bar alone, none once the size of the search was counted.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
PNA is the Pacific/North American pattern, a recurring shape in the North American jet stream. This tests whether its pattern shifts US west precip → western hydro utilities. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
QBO is the quasi-biennial oscillation, a reversal of stratospheric winds above the equator every couple of years. This tests whether its phase modulates hurricane counts. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; 1 of 2 cleared the usual bar alone, none once the size of the search was counted.
ENSO is the El Nino-Southern Oscillation, the Pacific climate cycle that swings between El Nino and La Nina. This tests whether salmon-vs-anchoveta-owner spread nowcasts ENSO before indices print. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
NAO is the North Atlantic Oscillation, the pressure seesaw that steers Atlantic storms into or away from Europe; and AO is the Arctic Oscillation, which governs how far cold air spills south out of the polar region. This tests whether ski-season snow (AO/NAO Dec-Feb) moves Vail. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
SAD is seasonal affective disorder, the winter dip in mood tied to daylight. This tests whether southern-hemisphere SAD: the same mechanism 6 months out of phase. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
TNA is tropical North Atlantic sea-surface temperature; and SST is sea-surface temperature. This tests whether its (tropical Atlantic SST) in May predicts hurricane-season insurer drag. Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
Tested at one and three months; nothing beat chance.
This tests whether full-moon vol (not direction). Checked across 2,862 occurrences in 14,260 trading days. Nothing separates it from chance.
This tests whether lumber demand proxies homebuilders 1-2 quarters early. Checked across 11,651 occurrences in 11,651 trading days. Nothing separates it from chance.
This tests whether lunar cycle: new-moon premium in equities (Dichev-Janes). Checked across 13,920 occurrences in 14,677 trading days. Nothing separates it from chance.
MJO is the Madden-Julian Oscillation, a pulse of tropical rainfall that circles the globe every month or two. This tests whether its phase modulates 2-4 week ag weather risk. Checked across 6,501 occurrences in 6,501 trading days. Nothing separates it from chance.
This tests whether options-expiry pinning. Checked across 665 occurrences in 14,260 trading days. Nothing separates it from chance.
This tests whether Payroll/FOMC calendar drift. Checked across 639 occurrences in 14,260 trading days. Nothing separates it from chance.
This tests whether post-hurricane OJ squeeze months. Checked across 44 occurrences in 6,235 trading days. Nothing separates it from chance.
This tests whether sea-level-pressure gradient trends → wind-generation capacity factors → utility mix. Checked across 7,605 occurrences in 7,605 trading days. Nothing separates it from chance.
This tests whether turn-of-month flows: returns concentrate in days −1..+3. Checked across 3,387 occurrences in 14,683 trading days. Nothing separates it from chance.
This tests whether friday-13th / superstition dips. Checked across 95 occurrences in 14,260 trading days. Nothing separates it from chance.
A deliberately meaningless test — geomagnetic Ap → gold as fear asset — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — month-length effect (23 vs 19 trading days) — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — reversed Halloween (long May-Oct) — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — TLT/duration placebo family for every new exotic signal — run beside the real ones to prove the screen can come back empty.
A deliberately meaningless test — zodiac-year returns — run beside the real ones to prove the screen can come back empty.
Cannot be evaluated on obtainable data. Requires IRI/BOM ENSO forecast plume archive.
Cannot be evaluated on obtainable data. Requires individual contract-month futures price series.
Cannot be evaluated on obtainable data. Requires historical option-implied volatility surface for agricultural futures.