Will the US stock market correct within the next year or two?

As of September 23, the S&P 500 closed at 7,706.03. A correction is conventionally measured from a subsequent market peak, not simply as a decline from that closing level. So the call is about the chance of a drawdown during the window, not a prediction that the index must be lower at its end. [1] [2]

Why the correction risk is elevated—but not a crash call

Recent history supplies a useful, imperfect base rate. MUFG counted 37 US equity corrections of 10% or more since World War II, including 13 declines of at least 20%, and estimated that 10% and 20% corrections occurred on average every 2.2 and 5.6 years, respectively, in its analysis published in November 2025. A crude constant-arrival-rate calculation using the 2.2-year interval implies roughly a 60% chance of at least one 10% correction in two years. That calculation is an illustration, not a calibrated forecast: it assumes a stable, memoryless frequency and says nothing about the drawdown’s timing, depth beyond 10%, or eventual recovery. [3]

The distinction between correction and bear market matters. Schwab notes that most corrections in its historical sample did not become bear markets; the sample is not a promise about the next episode. The fact that corrections recur is a reason to prepare for volatility, not to infer that a large bear market is overdue. [2]

The pressure point is the discount rate

The Federal Reserve raised its target range by 25 basis points on September 16, to 3.75%–4.00%; the statement said inflation remained elevated. On September 23, the 10-year Treasury par yield was 5.11%. Those rates do not mechanically cause a correction, but they raise the hurdle rate for future cash flows and make high-multiple equities more exposed to disappointment. [4] [5]

Inflation data explain why an easy policy pivot should not be assumed. August CPI was up 3.4% year over year, with core CPI up 2.4%; July PCE inflation was 3.7% year over year and core PCE 3.3%. The September Fed projections put 2026 PCE inflation at 3.7% and 2027 at 2.3%, but those are policymakers’ projections, not realized outcomes. If inflation stays sticky, longer rates could remain restrictive even if growth cools. If disinflation resumes, the same rate pressure could ease. [6] [7] [8]

Labor data are a counterweight, not proof of recession. August payrolls rose by 162,000 and unemployment was 4.1%; revised June and July gains were 31,000 and 21,000, respectively. The resulting three-month average was about 71,000 per month. That is slower hiring than the August headline alone suggests, but not in itself evidence that earnings are already collapsing. [9]

Valuation can turn good earnings into a falling index

FactSet’s September 18 report put the S&P 500’s forward 12-month P/E at 19.1—slightly below its five-year average of 19.8 and slightly above its ten-year average of 19.0. Analysts projected earnings growth of 31.8% for calendar 2026 and 15.2% for 2027. These are estimates, not reported future results; the forecast step-down in 2027 leaves less room for an earnings miss to be absorbed by growth. [10]

The composition of those forecasts also argues against treating the market as a single AI trade. FactSet expected all eleven sectors to report year-over-year Q3 earnings growth; Energy was expected to contribute the largest increase in estimated dollar earnings, while Information Technology was among the leading growth sectors. That breadth is a cushion, but broad index growth does not mean each constituent—or each shareholder—earns the same return. [10]

The S&P 500 is concentrated: its ten largest constituents represented 37.8% of index weight as of August 31, 2026. That makes index-level outcomes unusually sensitive to the earnings and valuation of a relatively small group of large companies. The weight is an exposure measure, not a forecast that those firms will underperform. [11]

The table isolates the arithmetic, not a forecast. It holds the starting forward P/E at FactSet’s 19.1 and applies assumed earnings changes and ending multiples. For example, 10% earnings growth paired with a decline to 17× would still produce about a 2% price decline; flat earnings at 17× would imply about an 11% decline. These are nominal price-return sensitivities, exclude dividends, and assume no change in index composition. They show why a correction can occur without a recession: multiple compression alone can offset earnings growth. [10]

AI is a real earnings engine—and a real cash-flow test

There is substantial realized demand in the supply chain. NVIDIA reported fiscal Q2 2027 revenue of $96.2 billion, up 106% year over year, including $89.0 billion of Data Center revenue, up 117%. That demonstrates strong supplier sales, not yet the ultimate return on every customer’s investment. [12]

Alphabet’s Q2 call described the other side of the spending cycle: capital expenditure was $44.9 billion, mostly technical infrastructure; quarterly free cash flow was negative $5.9 billion, although trailing-12-month free cash flow remained positive at $53.3 billion. This is an observed cash-flow consequence of investment, not evidence by itself that returns will be inadequate. [13]

The mechanism to watch is whether customers convert expensive compute capacity into durable incremental revenue and returns on capital quickly enough to support continued spending. A BIS analysis said the five largest hyperscalers were set to spend more than $1 trillion on AI-related capex across 2025–26; that figure is a forecast compiled from company calls and releases, not a realized final tally. Separately, a Johns Hopkins study points to grid-supporting equipment as a possible data-center bottleneck. Its unmet-demand figures are scenario outputs, not observed shortages. Delayed power or equipment could defer deployments and customer payback, even while chip suppliers report strong current sales. [14] [15]

Thus, AI exposure cuts both ways for the index: sustained customer returns can validate capex and support earnings, while weaker monetization or persistent infrastructure delays can challenge both growth expectations and the valuation assigned to them. This is an analytical risk channel, not a claim that an AI downturn is imminent.

What to watch—and what would change the call

The next scheduled FOMC meetings are October 27–28 and December 8–9, 2026; the December meeting is associated with new economic projections. The relevant signal is not simply whether the Fed cuts or holds, but whether inflation progress allows long yields to ease without a corresponding collapse in employment or earnings. [16]

Over the next several earnings rounds, watch revisions to 2027 estimates, reported free cash flow against capex, and evidence that new data-center capacity is being energized and used. The correction-risk view would weaken if inflation continues to cool, longer yields retreat, and earnings estimates hold up or rise. It would strengthen if inflation reaccelerates, long yields remain high or rise, and earnings revisions turn negative—particularly among the heavily weighted companies. A sustained widening in credit spreads would add concern; the September 23 high-yield option-adjusted spread was 2.73%, but one observation alone cannot establish a trend or rule out future stress. [17]

On balance: budget for a correction as a normal, plausible outcome over this horizon; do not treat it as a certain event or a reason to equate ordinary volatility with a bear market. The market can avoid a drawdown if earnings growth materializes and discount rates stabilize. Conversely, if both earnings expectations and valuation multiples reset, the same starting point can produce a much larger decline.