Signal Combinations Can Bracket a Bottom, but Cannot Pinpoint the Low
Contents8 sections
- Every signal class is either too early or too late, with a hit rate of about half
- The only stable pattern is the setup → divergence → confirmation sequence
- Three bear-market types need three different bottom-identification models
- Staged entry rules for long-term ETF investors
- Failure conditions must be written down in advance to be testable afterward
- 10/2026: no bottom signals lit, a top-type combination has formed
- Conclusion
- Frequently asked questions
The research conclusion is clear: the proposition that "no single signal can pinpoint the VOO/QQQ low in advance" holds, and no counterexample has been found so far. Every signal falls into one of two groups. The first group is ex-ante or coincident "setup" signals, such as breadth extremes, VIX spikes and extreme pessimism. They almost always appear near the low, but in multi-leg bear markets they often arrive 3 to 7 months early, and the index then falls another 25% to 40% after they trigger. The second group is ex-post "confirmation" signals, such as breadth thrusts, credit spreads peaking and rolling over, and the index reclaiming its long-term moving average. Their win rate is high, but the price is a lag of 8 to 15 trading days after the low (V-shaped) or even several months (bear markets), and by the time they confirm, the index has often already rebounded 10% to 15% off the low. The closest to "coincident" is the peak in HY credit spreads: in 2016, 2020, 2025 and 2026 it was less than one day away from the equity low. But a peak can only be identified after the fact, and in 2008 it came about 83 days before the final low. The workable approach is therefore not to guess the low, but three things: first classify the bear-market type (V-shaped crash, recession/credit type, bubble-burst/tightening type), then deploy in tranches along a "setup → divergence → confirmation" signal sequence, and write down the failure conditions in advance. As of 10/16/2026, the market is not in any bottoming zone. Not a single bottom signal is lit; what is lit are top-type signals: CAPE around 41, ERP around 1%, the Fed hiking, and extremely narrow breadth. If a bear market follows, the most likely type is bubble/tightening, which is exactly the type in which macro and breadth-thrust signals are least reliable for QQQ.
Every signal class is either too early or too late, with a hit rate of about half
First, define what "catching the low" means. This report breaks a signal into three testable attributes: the share of times it appears near the low (recall), whether it marks the actual final low when it appears (precision), and its time offset from the final low. Measured by these three yardsticks, the five signal classes fail in different ways, and the failures concentrate in the same scenario: multi-leg, recession or credit-driven bear markets.
Breadth and momentum setup signals have high recall and poor precision. In late 2008, only about 4% of NYSE stocks were above their 200-day line, and the S&P 500 then fell roughly another 25% before bottoming in 3/2009(StockCharts). Across 13 "washout" events since 2010, the median SPY return 60 days later was +3.6%, versus +3.7% for the full sample, so there is no visible edge over the 60-day window(thetrading.tools). The best structural evidence comes from new-low counts. In bear markets since 1929, with only one exception, the share of 52-week new lows at the final low was below the peak of the prior leg down; that is, "the low comes when new lows are contracting, not expanding." In 10/2008 more than 80% of stocks made 252-day new lows, while at the final low in 3/2009 only about 20% did(SentimenTrader). Divergences also show up at failed interim lows, however: 6/2022 produced a "striking positive divergence," and the market still broke to a new low in October(StockCharts DecisionPoint).
Breadth thrusts (Zweig, Deemer BAM, Whaley) are pure ex-post confirmation signals. In verifiable recent cases, they triggered about 8 to 15 trading days after V-shaped bottoms and lagged bear-market bottoms by 1 to 6 months. The "triple signal" of 1/2023 came 3 months after the 10/2022 low(Quantifiable Edges). Win rates after a trigger are high: across 18 ZBTs since 1945, 6-month and 12-month returns were all positive, averaging +15.3% and +24.0% respectively(Business Insider/Yahoo). But with a longer sample, McClellan counts 27 ZBTs since 1928, of which 12 were "bad" signals, and the major bottoms of 1940 and 1942 never triggered one at all(McClellan). This matters especially for QQQ: after the NYSE ZBT crossed 0.60, the Nasdaq-100's 12-month return was about -35% on both occasions, in 3/2002 and 2/2008(SentimenTrader).
Sentiment, volatility and credit follow the same pattern. The VIX closing peak was nearly coincident with the low in exogenous-shock selloffs: 7 days early in 2020, the same day in 2025, 3 days early in 2026. In 2008, however, it came about 109 days before the final low, and in 2022 the full-year high came in March, about 219 days before the month-10 low, without ever breaking above 40 that year(Macroption; convextrade). Based on 10 events, the research notes roughly estimate that "VIX peak equals the low (±10 days)" has a hit rate of about half. This is the researcher's own inference, not a source statistic. A VIX term-structure inversion has no directional edge by itself: from 2018 to 2026, the median SPY return 4 weeks after an inverted week was +1.45%, versus +1.69% after a contango week(FlashAlpha). Peak pessimism in sentiment surveys usually does not coincide with the low: AAII bears reached 60.9% on 9/22/2022, about 20 days before the low(AAII); the first "extreme fear" reading of the CNN Fear & Greed Index in 2022 came about 145 days before the low, and the subsequent 12-month return was only +2.9%(fffinstill); and the highest Put/Call reading on record came about 2 months after the 2022 low(StockCharts). Credit spreads perform best. The ICE BofA HY OAS peaked on 3/23/2020, 4/7/2025 and 3/30/2026, each less than one day from the equity low(FRED). The spread peak in mid-12/2008 came before the 3/2009 low, but when equities made a new low in March, credit did not make a new high, forming an actionable positive divergence(eco3min).
Valuation cannot be used for timing at all. CAPE at bottoms has ratcheted up from 6.6 in 1982 and 13.3 in 2009 to 27.1 in 2022 and 32.6 in 4/2025(multpl). Forward P/E explains an R² of only 5% of next-1-year returns, versus 31% for 5-year returns(JPM Guide to the Markets). At recession-type bottoms, trailing P/E can even point the wrong way: in 3/2009, because earnings collapsed, trailing P/E reached 110(multpl).
Macro and the Fed are lagging indicators overall. The S&P 500 low came on average 4.4 months before the NBER recession end (standard deviation 1.8)(VoxEU/CEPR), while the unemployment-rate peak came 4 to 9 months after the low(Fisher). The first rate cut came about 21 months and 18 months before the final low in 2001 and 2007 respectively, and the following 12-month returns were -13.5% and -20.6%(LPL). QE announcements were coincident with the low (9 days late in 2009, the same day in 2020); they mark the point where policy panic has hit its extreme and cannot be used to position early.
Price structure: "drawdown thresholds" offer almost no edge. From 1926 to 2022, buying at -20% or -30% produced 1-, 3- and 5-year returns "almost identical" to buying in an arbitrary month; only -40% or worse delivered excess returns, and there were only about 4 to 6 independent events of that kind(Of Dollars and Data).
Signal scorecard
The score (1 to 5) measures usefulness for marking the final low, not predictive power for long-term returns. "Effective sample" is an estimate after removing repeated readings within the same event.
| Signal | Nature | Relative to final low | Win-rate / hit evidence | Representative false signals | Effective sample | Data-snooping risk | QQQ applicability | Score |
|---|---|---|---|---|---|---|---|---|
| Very low share above 200/50-day line (<15%, <10%) | Ex-ante/coincident setup | V-shaped: coincident; multi-leg bear: months early | Post-2010 washouts show no 60-day edge | 10–11/2008, 7/2002 | About 13 (post-2010) | Medium | Medium | 2 |
| New-low contraction / positive divergence (on retest) | Semi-ex-ante, needs retest | Close to final low (2022: about 2 weeks after divergence) | Present in all bear markets since 1929 except one; 1-month win rate 18/21 | 6/2022 | About 31 in 93 years | Medium | Not verified for NDX | 3.5 |
| Breadth thrust (ZBT/BAM/Whaley/Nasdaq BT) | Ex-post confirmation | 8–15 trading days late (V-shaped) to 1–6 months (bear market) | Post-1945 12-month win rate about 90–100% | 1930s, 2011, 2015, 2022; NDX 2002/2008 about -35% | 18–30 | High (thresholds are in-sample fits, many definitions) | Low | 3.5 (VOO) / 2 (QQQ) |
| 90% up days | Ex-post confirmation | Within days | No public win rate | Three times in 10–12/2008 | Unknown | Medium | Low | 2 |
| VIX peak / >40 | Coincident, confirmed only after it falls back | About half within ±10 days; 109 days early in 2008 | 12-month median after VIX>30 about +18% (about 25 cases) | 2008 fell another ~25%; absent in 2022; 8/2024 | >40 about 6–8 times | Medium | Insufficient VXN data | 2.5 |
| VIX term-structure inversion | Coincident | Variable | No directional edge in 4-week returns | 6 recurring episodes in 2022 | 37 episodes | Low | Medium | 1.5 |
| AAII/II/F&G extreme pessimism | Ex-ante (zone) | Weeks to 5 months early | 12-month average +18% after AAII bears >+2SD (40 cases, 1987–2004) | 11/2008, 5/2022 | Few truly independent events | High (autocorrelation) | Medium | 2.5 |
| Put/Call extremes | Variable | Can be early or late (2 months late in 2022) | Qualitative only | 2007 top | Unknown | Medium | Medium | 1.5 |
| BofA Bull & Bear Indicator ≤2 | Ex-ante | Unknown | No buy-signal win rate found | Unknown | Unknown | Unknown | Medium | 2 (unverified) |
| HY OAS peak, and "new price low but spread not at new high" | Coincident confirmation | 2016/2020/2025/2026 ±1 day; 83 days early in 2008 | Consistent mechanism in recent events | 2002 may have lagged about 3 weeks (index definition to be checked); 2022 levels mild | About 6–7 | Low | Medium (mechanism is general) | 3.5 |
| CAPE / forward P/E thresholds | Zone, not timing | No fixed relationship | 1-year R² 5% | Single-digit CAPE 1974→1982 lasted 8 years; every bottom since 2010 had CAPE ≥20 | Few | High | Low (NDX data missing) | 1 (timing); 4 (long-term expected return) |
| EPS trough / estimates stop falling | Lagging | 1–3 quarters late | Price bottoms first (EPS estimates were still cut 16% after the 2020 low) | 2000–02 multiples kept compressing after earnings stopped falling | Few | Low | Low | 1 |
| Recession end, unemployment peak | Lagging, published after the fact | 4–9 months late; 11 months early in 2002 | — | 2000–02 | About 13 | Low | Worst | 1 |
| First rate cut | Conditional | 18–21 months early in 2001/2007 | 12-month median +18% without recession, -14% with recession | 2001, 2007 | About 11 | Medium | Worst | 1 |
| QE / policy-panic extreme | Coincident | 0 to +9 days | n=2 | — | 2 | Low | Medium | 2.5 |
| ISM trough, yield curve | Variable / danger zone | Order varies; bull steepening appears mid-bear | No recession followed the 2022–24 inversion | 2023 recession calls | Few | Medium | Low | 1 |
| Drawdown thresholds (-20%/-30%) | Ex-ante | After -20%, median further decline about 10%; in 3 of 13 cases another 35–46% | No excess return at -20/-30% | 2000–02, 2007–09 | -40% about 4–6 times | Medium (arbitrary thresholds) | Low (NDX once fell -83%) | 2 |
| Reclaiming 10-month MA / 200-day line | Ex-post confirmation | Weeks to months late | Cut the S&P maximum drawdown from -83% to -50% | Repeated whipsaws in V-shapes; no excess return 1940–2000 | Long, but parameters suspected of snooping | Medium | Medium | 3 (for risk control) |
Note: the further decline after -20% in the "drawdown thresholds" row was computed by the notes themselves from Yardeni's data on 13 postwar bear markets(Yardeni); "VIX hit rate about half" is also an inference. Both are derived figures that have not been independently verified.
The only stable pattern is the "setup → divergence → confirmation" sequence
Put the signal classes back on a timeline and the final low shows a recurring structure. Extreme panic (breadth washout, VIX peak, extreme AAII pessimism, surging new-low counts) marks the first leg of capitulation, not the low. The final low often comes on a later retest, characterized by "new price low, no new indicator low": fewer new lows, HY spreads not making a new high, and the share above the 200-day line no longer falling further. In June and September 2022 the share above the 200-day line was about 10% both times(Benzinga). Real confirmation comes from a subsequent breadth thrust or a resumed trend. Scott Brown (CMT) frames it as three steps: washout, divergence, thrust(CMT Association). BofA's Hartnett has similar signposts: Bull & Bear Indicator ≤2, the global breadth rule, capital capitulation, and a hawkish Fed peak(MacroBusiness).
This sequence has a different "compression ratio" in different bear-market types, which is why classification has to come first. V-shaped crashes (1987, 2020, 2025, 3/2026) have no retest, and the three stages are squeezed into 2 to 3 weeks: the extreme and the low coincide, and the thrust triggers within 8 to 15 trading days. 2026 is a textbook case: at the 3/30 low the HY OAS peaked the same day, the VIX peaked 3 days earlier, a Nasdaq breadth thrust triggered on 4/14(SentimenTrader), and the S&P 500 rose about 10.5% in April and set a new high(Moorman Harting). Yet no signal identified 3/30 in advance. Incidentally, sources disagree on the depth of this drawdown: -7.5%, about -8.4% and -9.1% are all cited. In the recession/credit type (2008–09) the sequence stretches to 4 to 5 months, and the first extreme comes early enough to cost 25% to 40%. In the bubble/tightening type (2000–02, 2022) the sequence can repeat several times; 2022 saw multiple "violent bear-market rallies" and failed thrusts: in 8/2022 the share above the 50-day line exceeded 90%, and the market still made new lows afterward(SentimenTrader).
There is an adversarial counterargument to face here: if confirmation signals have high win rates, why not simply wait for confirmation? The answer is asymmetric cost. At a V-shaped bottom, by the time the thrust arrives, a gain of about 10% to 15% is already gone. Hartford finds that about 76% of the best trading days fall in the first two months of a bear or bull market(Hartford Funds). Conversely, using only setup signals would have meant entering 4 months early in 2008. Vanguard's research shows lump-sum investing beats staged investing about 68% of the time(SmartAsset/Vanguard), so the value of staging lies in controlling tail risk, not in raising returns. This must be honestly listed as a cost of the rules.
Three bear-market types need three different bottom-identification models
Model A: V-shaped crash (exogenous shock, liquidity type)
Applicable conditions: the trigger is an exogenous shock (pandemic, war, tariffs, deleveraging); the decline is fast, reaching -15% to -20% within 100 days; HY OAS widens but no sustained credit-default cycle develops; employment and ISM were still healthy before the decline. The identification combination is "washout, plus HY OAS stops widening, plus a thrust within 10 trading days." Historical evidence is relatively strongest: bear markets that took less than 100 days were followed by an average of +19% over 6 months and +23% over 12 months, while slow bear markets came out "close to flat"(Schaeffer's). The trade-off: these declines are so fast that most of the capital should be deployed when the setup signals appear; waiting for confirmation usually means missing a rebound of 10% or more. The biggest failure risk is misclassifying the type, such as treating 10/2008 as a V-shape.
Model B: Recession/credit type
Applicable conditions: HY OAS has widened by more than 200bp within 90 days and is still widening; the unemployment rate is rising and initial jobless claims are climbing; the Fed is cutting rates or preparing QE. The core of identification is a credit divergence plus new-low contraction: when equities make a new low, HY OAS does not exceed its prior peak, and new-low counts are smaller than in the prior leg. Supporting conditions are a policy-panic extreme (QE or an emergency cut), and sentiment not becoming more extreme on the second test. Do not wait for the unemployment peak or the recession end; that would be 4 to 9 months late. The trade-off: the first extreme reading is bound to be too early (in 2008 the VIX, AAII and TED spread were all 3 to 5 months early), so deployment has to be stretched over several months, with the last tranche reserved for divergence confirmation. In 2002 the HY peak may have lagged the equity low by about 3 weeks, meaning even credit signals do not lead every time(Cambridge Associates). The index definition behind this data point still needs checking, so it is unverified.
Model C: Bubble-burst/tightening valuation-compression type (the key model for QQQ)
Applicable conditions: valuations were extreme before the decline (CAPE >35, very low ERP); the Fed is tightening because of inflation, or rate cuts fail to stop the decline; market leadership is a highly concentrated theme. 2000–02 is the extreme version: NDX -83%, taking 16.4 years to regain its prior high(ChartRow); the final low came about 11 months after the NBER recession end and about 21 months after the first rate cut. 2022 is the mild version: no recession, and the bottom came while the Fed was still hiking. In this type of bear market, both macro signals and NYSE breadth thrusts are unreliable for QQQ, because both NDX failures of -35% occurred in this setting. The identification combination should switch to "price structure plus NDX-specific breadth plus valuations back in a range compatible with the rate environment." Price structure means reclaiming and holding the 10-month moving average. NDX-specific breadth uses Arthur Hill's two-of-three rule: share above the 200-day line >60%, share above the 150-day line >70%, and the high-low spread >+10%(StockCharts). This rule only turned bullish in 2/2023, about 4 months after the 10/2022 low. The trade-off is accepting a clear lag in exchange for avoiding the "bear-market rally" trap. For long-term investors, in this scenario spreading purchases over time (24 months or more) is more robust than signal-based timing.
Staged entry rules for long-term ETF investors
The following rules apply only to new money or a set-aside "buy-the-dip reserve" and do not involve selling core holdings. All numeric thresholds are assumed parameters set from the evidence above; they are unbacktested rules and must be checked after the fact using the method in the next section. The reserve is split into four tranches of 25% each.
| Condition (if) | Action (then) | Basis and cost |
|---|---|---|
| S&P 500 drawdown ≥15% (QQQ ≥20%), and at least two of five washout conditions hold: VIX >35 or VIX/VIX3M >1, S&P share above 50-day line <15%, AAII bears >50%, a 5-day spike in Put/Call, HY OAS widening over 90 days but not above 600bp | Deploy tranche 1 | Setup signals are nearly coincident in V-shapes. Cost: months early in multi-leg bear markets |
| Drawdown ≥25% (QQQ ≥35%), regardless of signals | Deploy tranche 2 | Pure price rule, acknowledging that in 3 of 13 cases the market fell another 35–46% |
| The index shows a breadth thrust within 10 trading days (any of ZBT, BAM, Whaley or a Nasdaq thrust), while over the same period HY OAS is below its peak and the index has not broken its prior low | Deploy tranche 3 | Confirmation signal, postwar win rate about 90–100%. QQQ must additionally satisfy at least one of the NDX two-of-three conditions |
| S&P reclaims its 10-month moving average, and the share above the 200-day line is >50–60% | Deploy tranche 4 | Trend confirmation, lagging by 1–2 months or more |
| Type override B: HY OAS widened >200bp over 90 days and is still widening, or the unemployment rate rises consecutively | Tranches 1 and 2 are spaced at least 1 month apart; tranche 4 waits for "new price low but HY not at a new high, with new-low counts contracting" | Avoids the 2008-style "first extreme too early" |
| Type override C: CAPE >35 when the decline begins, or the Fed is still hiking | The QQQ portion switches to equal periodic purchases over 24 months; tranche 3 accepts only NDX-specific breadth, not NYSE thrusts | NDX fell about -35% after the 2002/2008 thrusts |
| The index breaks its prior low within 6 months after a breadth thrust triggers | Pause the remaining tranches and reclassify as type B or C | The failure mode of 2015 and 8/2022 |
| Drawdown below 15% | Do not touch the reserve; keep the existing periodic investments | -20/-30% by themselves show no excess return, and there is even less evidence for -10% |
Taiwanese investors face an additional currency factor (how the USD/TWD rate moves during US equity panics). These research notes do not cover it, so it is not included in the rules.
Failure conditions must be written down in advance to be testable afterward
The framework has three layers of falsifiability. The first layer is the core proposition: if some single signal, out of sample and with ≥10 independent events, lands within ±5 trading days of the final low in ≥80% of cases, with false signals (a new low within 6 months after triggering) ≤20%, then "no single signal can catch the bottom in advance" is refuted. The closest candidate today, the HY OAS peak, does not qualify: it can only be identified after the fact, n is about 6 to 7, and in 2008 it was 83 days early.
The second layer is the individual signal. Every trigger must record five things: the date, the distance from the final low in days and percent, the maximum adverse excursion (MAE) after the trigger, the 6-month and 12-month returns, and the bear-market type assigned at the time. Downgrade conditions include: a new low within 6 months after a breadth thrust triggers, a negative 12-month return after an NDX thrust, or a further decline of more than 25% after tranche 1 is deployed (meaning the type was misjudged). The researchers themselves acknowledge that the edge of thrusts "may have softened or even disappeared"(SentimenTrader), and Deemer has criticized peers for "torturing the data" to invent new thrusts(Walter Deemer). So only the few thrusts defined in advance are accepted, and no ad hoc variants are added.
The third layer is the model. After each full cycle ends (the index makes a new high), compare the staged rules with two benchmarks: lump-sum investing, and equal periodic investing over the same period. There are two comparison metrics: ending wealth and maximum drawdown. If for two consecutive cycles the staged rules' ending wealth trails lump-sum investing by more than about 2% (roughly the average gap estimated by Vanguard), without buying a clearly smaller maximum drawdown, signal-based staging should be abandoned in favor of pure periodic investing. Two data traps also need attention. First, some free breadth data backfill history using current index constituents, introducing survivorship bias(chartrow). Second, different vendors compute different ZBT dates; one reconstructed dataset has no signals in 2023 or 2025(thetrading.tools). The data source must be fixed when testing.
10/2026: no bottom signals lit, a top-type combination has formed
As of the 10/15/2026 close, the S&P 500 stood at 7,744, about -1% from the 10/6 record close of 7,818.93(Vista Partners; Digg). ChartRow lists the 10/6 high as 7,725, which contradicts FactSet's 10/9 quote of 7,765, so this report does not use it. The NDX closed at 29,935.6, about -4.1% from its record close(Investing.com).
| Category | Reading | Bottom threshold | Positioning |
|---|---|---|---|
| Breadth | S&P share above 200/50-day line 51.6%/44.1%; NDX 58.6%/50.0%; all US stocks above 200-day line 37.5% (1st percentile over 1 year)(thetrading.tools) | 50-day line <15% | No washout; the index is at highs but breadth is extremely weak, a top-type divergence |
| Constituent drawdowns | 84% of constituents ≥10% below their highs, 59% ≥20%(Motley Fool) | — | A "stealth bear market" inside the constituents, with the index propped up by concentration; the top 10 AI stocks make up about 40% |
| Volatility | VIX 16.46, VIX/VIX3M 0.89, contango for the 133rd consecutive day(thetrading.tools) | >35 or inversion | Calm; no VXN data |
| Credit | HY OAS about 3.16–3.23% (10/13–10/15, sources differ slightly)(MacroRadar; FRED) | Falling back after a peak | Near cycle lows, no stress |
| Sentiment | AAII bears 41.6% (average 31.5)(AAII); CNN F&G 31(Benzinga); Put/Call 0.88 neutral | AAII >50%, F&G <25 | Surveys lean fearful, but not extreme |
| Positioning | BofA Bull & Bear Indicator 8.1, still in the >8 sell zone(KuCoin) | ≤2 | Money has not de-risked; contradicts the fear in surveys |
| Valuation | CAPE about 41, highest since 2000; forward P/E 19.1, roughly the 10-year average; ERP about 0.94%(multpl; FactSet; Fortune) | Not applicable for timing | Expected-return range is low; ERP structure resembles 2000 and 2007 |
| Macro and Fed | 9/16 hike to 3.75–4.00%; 10-year yield about 5.1–5.3%; CPI 3.7%, PPI 6.3%; ISM 54.5; nonfarm payrolls +29,000, unemployment rate 4.2%(Federal Reserve; First Trust; Fox Business) | Rate cuts, QE, ISM trough | Mid hiking cycle; the macro bottoming sequence has not started |
| Price structure | Both indices near prior highs; the deepest 2026 drawdowns were in March (S&P about -7.5% to -9%) and month 7 (Nasdaq Composite -10.1%) | Drawdown ≥15% | No tranche in the rules table has triggered |
Several figures conflict and call for reserved judgment. The real-rate basis for ERP is contradictory: Fortune uses a 10-year TIPS yield of 2.86%, while thetrading.tools puts the real rate at about 1.5%. Core inflation also has two sets of numbers: CPI up 2.6% year over year, versus the Fed's projected core PCE of 3.4%. The NDX forward P/E of about 23 comes from a single source, Siblis(Siblis). No public current reading of the Goldman bear-market indicator was found; "possibly above 60–70%" is only an inference from its inputs.
Positioning judgment: under the rules table, none of the four tranches triggers, the reserve stays untouched, and existing periodic investments continue. More noteworthy is that if a bear market follows, the combination of conditions fits Model C almost perfectly: CAPE 41, ERP about 1%, the Fed hiking because of inflation, doubts over AI capex and its financing, and the top 10 stocks at about 40% of the index. This implies two things. First, a "Fed rescue" would come much later than in 2020, because inflation has tied the Fed's hands, a situation similar to 1974 and 1981. Second, for QQQ, the two most often cited classes of "bottom-catching signals," NYSE breadth thrusts and macro turning points, are precisely the ones that have historically failed worst in this type. If AI financing doubts spread to credit markets and HY OAS starts widening sharply, the pacing limits of Model B should be layered on top.
Conclusion
What this research changes is not whether the low can be caught; the answer was always no. What it changes is the recognition that the cost structure of being wrong shifts with the bear-market type. In a V-shaped crash, waiting is the most costly; in the recession/credit type, being early is the most costly; in the bubble/tightening type, trusting macro signals and breadth thrusts is the most costly, and for QQQ holders that is an existential risk. So what really needs preparing in advance is not signal thresholds but type classification: valuations, the Fed's direction and credit conditions at the start of a decline determine which rule set to use better than any extreme reading during the decline.
The second non-obvious inference is that the most informative reading right now is not in the bottom-signal table, but in the contradictory combination of "index at highs, eight in ten constituents in pullbacks, and positioning still crowded." It does not predict when a decline will happen, but it has already determined what kind of decline it would most likely be. The most actionable step for long-term ETF investors is to write down Model C's QQQ time-diversification rules and failure conditions now, while no panic is clouding judgment.
Frequently asked questions
Is there a single signal that can catch the VOO or QQQ low in advance?
Based on the evidence compiled in this article, no. Ex-ante or coincident setup signals often arrive 3 to 7 months early in multi-leg bear markets, while ex-post confirmation signals lag the low by 8 to 15 trading days or even several months. The HY credit-spread peak, the closest to coincident, can also only be identified after the fact, and in 2008 it came about 83 days before the final low.
Why classify the bear-market type first?
Because the same set of signals has very different timing offsets in different bear markets. In a V-shaped crash the three stages may be squeezed into 2 to 3 weeks, in the recession/credit type they may stretch to 4 to 5 months, and in the bubble/tightening type they may repeat several times. If the type is misjudged, the staging pace may end up too early or too late.
Are breadth-thrust signals reliable for QQQ?
Not very, in bubble-burst/tightening bear markets. After the NYSE ZBT crossed 0.60, the Nasdaq-100's 12-month return was about -35% on both occasions, in 2002 and 2008. That is why, in the type C scenario, this article's rules accept only NDX-specific breadth for QQQ's tranche 3.
Isn't it safer to wait for confirmation signals before entering?
Confirmation signals have higher win rates, but the cost is asymmetric: at a V-shaped bottom, a gain of about 10% to 15% may already be gone by the time the thrust arrives. Conversely, using only setup signals would have meant entering 4 months early in 2008. The staged rules in this article are a trade-off between the two, and they are unbacktested assumed parameters.
As of 10/16/2026, is the market in a bottoming zone?
Based on the readings in this article, no. Not a single bottom signal is lit, and none of the four tranches in the rules table has triggered; what is lit is a top-type combination such as CAPE around 41 and ERP around 1%. This is a positioning judgment from research notes, not investment advice.