Bullish Percent Index Near Oversold With SPY Near a Record: Seven Previous Cases

Research note. Computed from our daily Bullish Percent Index since March 10, 2010and TradeStation daily closes for SPY, RSP and six sector and size ETFs, all frozen at the September 18, 2026 close. The methods are in How we checked it below; our editorial and AI standards are in How we use AI.

• ·Near oversold, precisely. Our Bullish Percent Index closed at 30.6%. The conventional oversold line is a reading below 30. This study uses 35 or lower, because under 30 with SPY near a record has happened on six sessions since 2010.
• ·Rare. SPY is 2.1% below its record. Breadth at or below 30.6% with the index that high has occurred on 11 of 2,125 near-record sessions, or 0.5%.
• ·Seven cases, or five. Six of seven were higher a month later. A quarter later three were, with a median of −1.8%, and four became 10% corrections within six months. Spaced at least six months apart, five cases remain and their quarter looks ordinary: a median of +2.6%, three higher.
• ·The reputation belongs to the correction. The same reading after a 5% decline was followed by +4.7% a quarter later. So was the first close more than 5% below a record, whatever the BPI read.

On August 14, 3,035 of 4,795 US stocks were on a point-and-figure buy signal. On September 18 it was 1,460 of 4,777. That is 1,575 fewer charts saying buyers are in control, in five weeks over which the S&P 500 ETF lost 1.9% and closed 2.1% under its record.

That count, as a share of all stocks, is the Bullish Percent Index, and at 30.6% it sits a hair above the level chartists have called oversold since Abe Cohen built the indicator in the mid-1950s. StockCharts’ reference page gives the convention as overbought above 70 and oversold below 30. Our aggregate is not there; the NYSE-listed half of it, at 27.9%, is. So: near oversold. A sub-30 reading with SPY near a record is too rare to grade, and the study below uses 35 or lower. The rule handed down with the indicator is that a washed-out reading is where you start looking for the turn. It skips a question, which is what the index was doing while breadth washed out.

Washed out is the normal state of a correction

Look at where the line dips into the shaded zone: 2011, the 2015 to 2016 slide, late 2018, March 2020, long stretches of 2022, autumn 2023, April 2025. Those are stretches you remember because the index was falling. A low BPI is usually the breadth-side picture of a correction everyone can already see in the price. When SPY has been within 3% of a record the typical reading is 59.4%.

So there are two ways to arrive at a washed-out BPI. In one, the index has dropped more than 5% from its high and most stocks have gone with it; that has started 30 episodes since 2010. In the other, the index is within 3% of its high and the stocks have gone without it; that has started eight, and Monday, September 14 was the eighth. SPY’s forward return depends on where it stands against its record whatever breadth is doing, so each gets its own baseline.

Start with the lower block, because it is where the indicator’s reputation comes from. A BPI at 35 or lower with SPY already more than 5% down was followed by +4.7% over the next quarter, against +3.1% for all sessions. That looks like an edge until you ask what any session that far below a record returned: +4.1%. A nearer comparison is the first close more than 5% below a record, whatever the BPI read. There have been 19, and they returned +4.7% with a median of +6.4%, 15 of them higher. Eight of those 19 days already had the BPI at 35 or lower; the other eleven averaged +3.9% with a median of +6.7%. Since 2010, buying after a 5% decline has paid whatever the BPI said. I cannot find an edge in what is left over.

The top row is the one we are in, and its baselines point the other way. Sessions within 3% of a record returned +2.2% over the next quarter with a median of +3.3%, higher 76% of the time. Sessions 1% to 3% below a record, where all eight of these episodes began, did better still at +2.9%. Against that, the seven completed cases averaged +0.2% with a median of −1.8%, and three of the seven were higher. The month in between looks fine. It is the quarter that went missing.

What happened in the 7 previous cases?

Seven is a short list, short enough to read line by line.

Three of the seven ended without much damage. November 2016 was a pre-election slide, and SPY was at a new record ten sessions later. July 2021 was a five-session scare. September 2014 took a 7.7% detour first, the October Ebola-and-oil slide, and the quarter still finished up 5.4%.

In the other four the index caught down to the stocks. July 2015 led into the August crash, and SPY did not see a new record for a year. December 2015 lost 9.6% in the following month. December 2021 and December 2024 are the pair I find hardest to dismiss, because both looked benign at first: SPY made a new record within 6 and 20 sessions, and the one-month return was +4.4% and +2.8%. At its worst over the following six months SPY closed 17.1% and 18.3% below the record it had started under.

Now the part that cuts against the scary reading. December 2015 began 111 sessions after July 2015, and December 2021 began 97 sessions after July 2021. For the six-month columns that matters directly: the paired windows share 15 and 29 sessions. For the quarterly return it does not, since no two 63-session windows touch. Even so, a stricter rule that requires six months between cases is a fair sensitivity check, and its effect is large. Five cases remain: a mean of +1.7% over the next quarter, a median of +2.6%, three of five higher, two of five turning into 10% corrections. That is close to an ordinary near-record quarter. Seven cases look weak, five look unremarkable, and the conclusion I can defend is the gap between them: the record here is too thin to lean on in either direction.

Laid over each other the seven paths look like a forecast plume, and I would resist reading them as one. What the picture does show is how little the first month settles. Six of seven were higher at session 21, and even then the outcomes ran from −9.6% to +5.4%. By six months they ran from a 10.2% loss (December 2021) to a 12.5% gain (November 2016), with the average 1.4% above the starting close against 4.8% for an ordinary near-record session.

Is a breadth recovery the tell? A hypothesis, with its weakness

Looking for what separated the cases, one thing stands out in the lower pane. In September 2014, November 2016 and July 2021 the BPI closed at 65 or higher within 42 sessions. In the other four it did not, and those are the four that fell 10% below their record. A perfect split, and I do not trust it. I found the 65 line by looking at these seven cases, so it has been fitted to them and tested on nothing.

It also flatters itself on timing. Nobody knows on day one whether breadth will recover. You know at session 42, and graded from there the split is much less clean.

The three recoveries returned −0.7%, +5.9% and +3.7% from session 42 onward and never lost more than 4.5%. Of the four non-recoveries, both 2015 cases had taken most of their losses by then: July 2015 lost a further 1.5% and December 2015 gained 7.5%. The other two, December 2021 and December 2024, went on to lose 14.8% and 16.5% at their worst. So the usable version of the idea is modest. When breadth had not recovered by session 42, two of four cases still had a large decline ahead of them, and when it had, none of three did. September 2026 is the first case this can be tested on without hindsight.

Who is doing the falling

The mechanics this time are easy to see. An index weighted by size can hold its level on its largest members while the median stock slides, and that is what the last five weeks look like. From the August 14 breadth peak to the September 18 close, SPY lost 1.9%. The equal-weight version of the same 500 companies, RSP, lost 4.7%. Technology (XLK) was flat and energy (XLE) rose 3.9%, while industrials (XLI) fell 9.0%, utilities (XLU) 7.2%, small caps (IWM) 6.9% and consumer discretionary (XLY) 6.1%. Only 31% of stocks closed that Friday above their 50-day average and 43% above their 200-day. New 52-week lows outnumbered new highs 267 to 51.

This is the third time in three months we have written about the same shape from a different angle. In July, nearly half the Nasdaq-100 was in its own bear market with the index 3% from a record. In August the question was why only tech was going up, and the sector-level history said narrow leadership near highs was not a sell signal on its own. I still think that. Narrow leadership describes who is winning. What changed in September is that the rest of the market stopped treading water and started breaking down on its own charts, which is the thing a point-and-figure count measures and a sector spread does not.

How I would use this, and the test that starts now

I would not buy this reading as an oversold signal, because the evidence that it is one comes from corrections and mostly belongs to them. I would not sell stocks because of it either; five to seven cases cannot support that. What I take from it is narrower. Six of the seven cases were higher a month later, including the two that did the most damage afterwards, so a rally from here would tell me nothing about which kind of case this is. The uncomfortable part is that the two worst outcomes were the two that felt safest a month in.

So here are three markers and one outcome test, written down before the outcome, each with a rule and a deadline.

Marker A is the recovery hypothesis with a deadline it cannot move, and the outcome test is what scores it: a recovery followed by a 10% loss from the session-42 close is a failure, and I will say so here. Marker B moves us into the lower block of the first table, where a 5% decline has been worth buying with or without the BPI, and it overrides the other two. Marker C describes December 2021 and December 2024, and its record is mixed. Applied to the seven past cases it would have fired three times. Twice, on January 10, 2022 and February 24, 2025, SPY went on to lose more than 16% from that day. Once, on August 19, 2021, it fired at session 24, breadth then recovered to 65 by session 34, and SPY gained 6.4% over the next quarter. So C can fire before A is decided. If both are met, A takes precedence and C goes down as a false alarm, as it would have in 2021. Two out of three, fitted after the fact, is a reason to pay attention and nothing more.

I am confident the divergence is real and rare. How it resolves is the part I hold loosely.

How we checked it

Every number here comes from our own Bullish Percent Index and the daily closes of nine ETFs. In plain terms:

• ·The Bullish Percent Index is the share of stocks whose point-and-figure chart last gave a buy signal (a column of rising prices exceeding the previous one) rather than a sell signal. We chart every US common stock in our price database with a 1% box and a three-box reversal, about 4,800 names today. It is our own series. StockCharts and other vendors build theirs from index memberships and traditional box sizes, we have not measured how closely ours tracks them, and every threshold here applies to ours alone. The study starts on March 10, 2010, the first session with at least 2,000 stocks in the count.
• ·Why 35 when oversold means below 30. The conventional line is below 30. Near a record, a reading under 30 has occurred on 6 sessions in the whole study, too few to grade, so the cut is 35 or lower and the piece says “near oversold”. Other cuts, next-63 mean and median: BPI ≤ 32, within 3%, 4 cases, -0.4% and -0.6% (2 higher); BPI ≤ 35, within 3%, 7 cases, +0.2% and -1.8% (3 higher); BPI ≤ 40, within 3%, 14 cases, -0.2% and -0.1% (7 higher); BPI ≤ 35, within 5%, 13 cases, +2.8% and +2.6% (8 higher). Widening the index condition to 5% pulls the quarter back toward the baseline.
• ·Distance from the record is SPY’s close against its highest close on or before that day, over its whole history since 1993, so a 2013 session is measured against the 2007 high. “Near a record” is within 3%. “After a correction” is more than 5% below. Sessions between 3% and 5% are in neither block.
• ·An episode starts on the first day its conditions hold after 21 sessions without such a day, so a three-week washout counts once. Blocks are clustered separately, which means one washout can appear in both if the index recovers faster than breadth: December 2015, December 2021 and December 2024 first crossed 35 with SPY more than 3% down and entered the near-record block as the index rebounded. The first-close row in the correction block applies the same rule to the index alone: the first close more than 5% below a record after 21 sessions without one. It is a true first crossing, and 8 of its 19 days also had the BPI at 35 or lower, so it overlaps the washout row rather than excluding it; the other 11 averaged +3.9% with a median of +6.7%.
• ·Spacing. A case is marked “overlaps” when it begins within 126 sessions of the previous case that was kept. At that spacing the 126-session windows overlap and the 63-session windows do not, so the five-case row is a more conservative counting rule whose effect we disclose. It does not correct double counting in the quarterly figure, and nothing here establishes that the remaining five are independent.
• ·Returns and losses. Forward returns are SPY price changes, without dividends, graded only where the window is complete. “Worst loss from entry” is the lowest close of the window against the starting close; it is not a maximum drawdown, which would measure from the highest close inside the window and is larger in five of the seven cases. The six-month column is a third measure: the lowest close of 126 sessions against the record in force on day one. The 23% base rate is the share of all near-record sessions after which SPY closed at least 10% below that day’s record within 126 sessions.
• ·What is frozen and what updates. Our BPI is rebuilt every day from the current stock list, so a past reading can move when a company is delisted or a price is corrected, and a moved reading can change which days qualify. The study tables, baselines and historical paths are therefore rendered from a snapshot saved at publication (Sep 20, 2026), with a SHA-256 fingerprint of each of its 11 input files, and it is available for download below. The latest-reading box, the chart lines and the marker panel read current data. If a recomputation ever disagrees with the snapshot in a way that changes a conclusion, we will restate it on the page and bump the modified date.
• ·Survivorship. The BPI is computed from stocks that are listed today. Our archive of delisted symbols is not read by this calculation, so companies that failed or were acquired are missing from every past reading. We have not measured the size of that gap. It can shift the level of the line, and because episodes are picked by threshold crossings it can also change which days qualify and therefore which cases appear in this study.
• ·The BPI Manual runs a related study (readings under 30, regardless of where the index stood) and finds above-baseline returns against all sessions. Of the 24 sub-30 episodes in this study’s window, none began with SPY within 3% of its record. They belong to the lower block of the table, where the correction baseline explains most of the return.