19 Midterms, 19 Wins: Election Day 2026 Tests the Record With a Hiking Fed, $104 Oil and Debt-Funded AI Capex

Measured from the October close in each of the 19 midterms since 1950, the index was higher 12 months later every single time, for an average gain of 16.6% and a median of 16.3%. Measured from the same October close to year-end, it was higher in 13 of 19 for an average of 3.6%. The December window has failed six times in nineteen: 1974 (-7.2%), 2018 (-7.6%), 1994 (-2.4%), 1986 and 2022 (-0.8% each) and 2002 (-0.7%).

The mechanism is easy to state and important to keep straight. The 12-month record is not a claim about November, and the consensus midterms trade buys it as if it were. What the data actually says is that the index reliably ends a midterm year higher while the run into that year-end is a coin flip with a fat left tail, and that the fat left tail is systematically fatter when the Fed is tightening into the vote. A Financial Times analysis of Robert Shiller's data puts the average peak-to-trough decline inside a midterm year at 18% since 1942, against a 9% deepest fall so far in 2026, and the cyclically adjusted P/E at 40.7 versus a previous midterm high of 33.8 and a long-run average near 20.

A hiking Fed on Election Day is not unprecedented. CME Group's study of 27 federal election years from 1972 finds the Fed took no action between Sept. 1 and Election Day in 52% of them, and that it raised rates near a midterm four times — 1978, 1982, 2018 and 2022 — every instance tied to rising inflation. It acted in November immediately before an election only twice, both midterms, in 1978 and 2022. A December 9 hike would repeat that pattern almost exactly, with one crucial difference: this time the inflation impulse is a supply shock from an active Iran conflict rather than a domestic demand overshoot.

The mechanics of the election compound the policy problem rather than offsetting it. Republicans defend a 218-214 House majority and a 53-47 Senate majority, and the Democratic path runs through 11 Senate races that decide control, so the result may not be settled by the time the FOMC votes. The market has largely priced the outcome that history says is weakest for stocks: when Democrats held both chambers and a Republican held the presidency — 1986, 1990 and 2006 — the reaction day averaged -0.5%, the one-month return was +3.4% and the one-year return was +10.9%, against +13.7% for midterms since 1980. A Democratic sweep is priced at 76%; the configuration with the best historical record, divided government, is priced at 19% and has averaged +15% over a year, positive in 14 of 15 elections.

This is the part the 19-for-19 record has never had to absorb. The hyperscalers no longer fund most of their data centres out of operating cash flow. Incremental debt has moved from 9% of capital spending in FY2024 to 32% on a trailing basis through June 2026, and aggregate debt across the group is roughly $700 billion, with about $250 billion of 2026 capex expected to be debt financed. The marginal AI dollar is priced off the same policy rate the Fed is raising, which means the AI buildout now carries interest-rate risk it did not have in any prior midterm cycle.

The Bank for International Settlements describes the structure in detail: special-purpose vehicles and joint ventures acquire or develop the data centres, are capitalised by a consortium of sponsors and raise debt through private placements, while the hyperscaler keeps a minority stake and supplies long-term operating leases, capacity offtake agreements and guarantees. Debt is serviced by lease cash flow and held by private credit funds and other institutional investors, and banks support the vehicles with funding lines. That last link is the transmission channel: private credit appetite, refinancing pressure at the vehicle level and guarantee activation are where a policy rate becomes a capital expenditure decision.

That last point is the loop that makes 2026 different from every prior midterm. The equity being sold to fund the buildout is priced off the same discount rate the Fed is raising, the debt being issued to fund it is priced off the same curve, and the customers whose budgets justify it are increasingly the leveraged entities. Nothing in the 1950-2022 record covers a Federal Reserve tightening into a vote while the dominant source of index earnings is being financed at a 32% debt share on other people's balance sheets.

The demand chain has not cracked, which is precisely why the credit structure matters. NVIDIA grew revenue 106% to $96.2 billion in its June quarter with data center revenue of $89.0 billion, up 117%, at a 75.0% gross margin. TSMC grew third-quarter revenue 50.9% to NT$1.49 trillion at a 66.4% gross margin. SK hynix grew revenue 328% to KRW104.6 trillion with operating profit of KRW83.3 trillion, an 80% margin, as HBM4 shipments began in earnest, and Tokyo Electron grew revenue 48% to ¥935.1 billion with operating profit up 86% to ¥295.4 billion, guiding full-year operating profit above ¥1 trillion for the first time. Every one of those four is paid on volume, not on the customer's credit.

The president has warned that a Democratic House win would bring 'an economic depression, much like 1929', while pledging $5,000 per adult citizen if Republicans retain Congress. Whether that rhetoric is priced as a warning about the opposition or a promise about one's own majority, both branches expand the deficit in a market where the 10-year yields 4.92% and the 30-year 5.29%, with 10-year yields already up roughly 100 basis points since 2024 on the inflation and monetary-policy effects of an AI-stoked economy, tariffs and Iran-fueled oil shocks. Reuters columnist Mike Dolan argues the biggest market impact of these midterms may not arrive until the 2028 countdown begins. The midterms are therefore a bond-market event first and an equity event second.

The composition matters for how the tape behaves in the days after the vote. A Republican Senate and Democratic House is priced at 19% and has historically produced the strongest equity record, +15% over a year on average and positive in 14 of 15 elections. A Democratic sweep at 76% maps to the weakest configuration in the sample at +10.9% over a year. A Republican sweep at 6% restores full single-party control, which has averaged +11.5% over a year. With results in 11 Senate races potentially unsettled for days and a December 8-9 FOMC meeting inside the same window, the path risk in the 0.0% average of the prior hiking-Fed midterms is the more relevant statistic than the 16.6% twelve-month average.

The consensus midterms trade buys an index because the S&P 500 has never failed to be higher a year after a midterm. That record has already survived four hiking-Fed midterms, so the endpoint is the wrong thing to worry about; the December window, where those four averaged 0.0%, is the right one. What makes 2026 harder than 1978, 1982, 2018 and 2022 is not the rate hike itself but that the same policy rate now reprices the credit funding a $700 billion AI buildout, at a cyclically adjusted multiple of 40.7 versus 33.8 at the last midterm peak. Own the suppliers paid on volume and the producers paid on the barrel, and be skeptical of the buyers funding the buildout on other people's balance sheets.