top of page
  • Instagram
  • X

S&P 500 RSI at 24.69: A Recovery Signal, Not a Floor

10 hours ago
23 min read

What the oversold chart actually establishes

 

S&P 500 RSI at 24.69: A Recovery Signal, Not a Floor

 

The S&P 500's reported 14-day Relative Strength Index closed at 24.69 on November 20, 2026. The accompanying historical comparison identifies this as the first close below the conventional oversold threshold of 30 since March 20, when the indicator registered 29.88. That is a meaningful description of recent price pressure, but it is not a statement that the index has reached fair value, exhausted every seller or secured a positive return from the next opening price. The useful question is narrower: how much should this observation change an investor's assessment of future returns, and what uncertainty must travel with that change? A disciplined interpretation starts with the actual statistics rather than the emotional resonance of the word oversold.

The reported backtest uses daily S&P 500 closing prices from January 1970 through November 2026 and describes 41 readings below 30. Average subsequent returns were 4.2% after three months, 7.6% after six months and 11.6% after 12 months. The corresponding averages across all periods were 2.3%, 4.7% and 9.6%. Subtraction gives conditional advantages of 1.9, 2.9 and 2.0 percentage points, respectively. These are horizon-specific differences in reported arithmetic averages, not annualized excess returns and not the performance of a self-financing trading strategy. In particular, one cannot multiply the three-month advantage by four and present that result as a demonstrated yearly premium. Doing so would require repeatable opportunities, an explicit capital-reinvestment rule and evidence that the opportunities do not overlap.

The lower panel adds an important dimension that the return averages alone conceal. Signal periods were positive in 80% of the three-month and six-month observations. At 12 months, however, only 74.4% of the 39 completed windows were positive. The plotted all-period hit rate at that final horizon is visibly higher than the signal hit rate, even though the signal's average return is higher. These two results can coexist because the magnitude of winners and losers matters as much as their frequency. A strategy can win less often but win more in its successful cases, or suffer fewer small losses while remaining vulnerable to a large one. Neither the mean nor the hit rate determines the other.

The top panel also cautions against translating a medium-horizon pattern into immediate timing confidence. It shows an approximately flat relationship at two weeks, an annotated 0.9% signal return at one month, and a more substantial separation between signal and baseline at three and six months. The lower panel does not display a uniform hit-rate advantage at every horizon either: the lines converge around one month and the signal line is slightly lower at two months. Values without explicit labels should not be read as precision estimates from the pixels. The broad message is sufficient: the visually strongest evidence concerns a recovery over months, not an assured rebound during the next session. A connecting line between discrete horizons is only a presentation device, not a forecast of the path between them.

There is a methodological inconsistency that deserves to remain visible. The descriptive text gives a January 1970 start, whereas the chart footer says the backtest begins in 1971. Without the underlying price series and event ledger, it is not possible to determine whether this reflects a labeling error, an initialization convention or genuinely different samples. The analysis here therefore treats the numbers as reported statistics rather than an independently replicated result. The discrepancy does not establish that the return comparisons are false, but it limits confidence in their exact reproducibility. A one-year difference can matter when the sample contains only 41 selected events, particularly if one or more early episodes materially affect the average.

There is a second definitional issue. The phrase “41 readings” sounds like a count of individual daily observations, while the chart describes “41 signals.” Those are not automatically equivalent. A market can remain below 30 on consecutive days, generating numerous daily readings during one selloff. A study might instead select the first downward crossing, require a reset above the threshold, or impose a waiting period before admitting a new event. Each choice produces a different experiment and a different practical entry rule. The displayed totals alone do not identify the rule. It would therefore be inappropriate to assert that buying every sub-30 close has produced the reported outcomes. Event construction is part of the economic claim, not merely a footnote for programmers.

The reported maximum drawdown of 14.3% within signal windows is equally material. It prevents the attractive averages from being mistaken for an absence of interim loss. The exhibit does not fully specify whether this number is measured from entry, from an intervening peak, across individual windows or on a composite strategy. Those distinctions change how a portfolio manager could use it. What can safely be said is that the reported favorable forward returns coexisted with significant adverse movement. It is not safe to interpret 14.3% as a future downside ceiling, a guaranteed stop-loss level or the loss associated with the typical event. A historical maximum depends on the sample and on the measurement convention.

Taken together, the chart supports a conditional recovery hypothesis with several boundaries. The relevant benchmark is not zero return but an equity market that already had positive average returns in the full sample. The strongest apparent improvement is not necessarily at the longest horizon. The 12-month observation count differs from the headline signal count, and the event definition and starting date need reconciliation before replication. None of those qualifications requires dismissing the signal. They determine the appropriate weight to give it: evidence that unusually weak recent price action has sometimes preceded favorable medium-term outcomes, rather than proof that a specific November close is an investable bottom. The rest of the argument must explain the mechanism, test the statistical uncertainty and connect the result to a portfolio capable of surviving the path.

 

RSI measures the path, not the value of the market

Relative strength in RSI is an internal comparison between recent gains and recent losses in the same instrument. It is not relative performance against another index, a measure of earnings strength or an estimate of expected return. Under the familiar Wilder convention, positive and negative daily price changes are separated and smoothed. If G denotes the smoothed positive change and L the smoothed absolute negative change, the indicator can be written as RSI = 100 - 100 / (1 + G/L). Its bounded scale makes different market episodes easy to display together, but the transformation does not create additional fundamental information. Everything entering the calculation comes from the chosen price history.

For a 14-day Wilder calculation, the recursively updated gain estimate is G_today = (13 × G_yesterday + gain_today) / 14, with an equivalent update for losses. The initial seed is commonly based on a simple average of the first 14 observations. This explains why “14-day” does not necessarily mean that all earlier information disappears after exactly 14 trading sessions: recursive smoothing leaves a declining influence from previous observations. Different initialization methods, simple rolling averages and alternative data adjustments can create different readings near a threshold. An exact replication of 24.69 therefore requires the same price series, calculation convention and sufficient warm-up history. Merely selecting a platform's indicator with the same name is not a complete replication protocol.

The reported level can nevertheless be translated into a useful mechanical relationship. Rearranging the formula gives G/L = RSI / (100 - RSI). At 24.69, this ratio is approximately 0.328. At the threshold of 30, it is approximately 0.429. At the prior reported value of 29.88, it is approximately 0.426. These numbers describe the balance of smoothed positive and negative price changes, not the probability of an up day or the fraction by which equities are undervalued. A lower reading says that recent losses have dominated gains more strongly under the indicator's weighting rule. It does not say that losses must now stop, nor does it quantify how much cash is waiting to buy.

A simple hypothetical illustrates the limitation. Consider two markets with the same proportional balance between smoothed gains and losses, but with daily moves twice as large in one market as in the other. Their RSI values can be identical even though their volatility and the dollar losses borne by holders differ considerably. Likewise, an index near a low RSI might still trade at a demanding valuation if its earlier starting price was exceptionally high. A sharp fall in the numerator of a valuation ratio does not guarantee cheapness when the denominator, expected earnings or sustainable cash flow, is also falling. RSI describes a path through prices; valuation requires an assessment of the claims those prices purchase.

The threshold itself introduces a discontinuity in the trading rule that need not exist in the economy. A reading of 29.99 and a reading of 30.01 are nearly indistinguishable economically, yet a binary oversold system may treat them as opposite states. The reported November value is well below that boundary, but this does not remove the broader issue. Researchers should examine neighboring thresholds and continuous specifications to see whether the result is robust or depends on a narrow cutoff. Testing many alternatives after seeing the returns creates its own selection problem. Sensitivity analysis is most useful when it asks whether a mechanism survives reasonable variations, not when it searches for whichever threshold produces the prettiest historical line.

There are also two plausible but competing ways to interpret a weak price path. Temporary selling pressure can push prices below levels that patient capital is willing to pay once balance-sheet capacity returns. In that case, a low RSI may coincide with a reversal opportunity. Alternatively, successive negative returns can represent the gradual incorporation of worsening information. Earnings revisions, financing stress or a change in the discount rate may persist rather than reverse. The indicator does not identify which process generated the decline. The same statistic can therefore be associated with an attractive liquidity concession in one regime and an incomplete fundamental repricing in another. Calling both regimes oversold does not make them economically identical.

This tension is consistent with the coexistence of different return dynamics across horizons. Moskowitz, Ooi and Pedersen's research on [time series momentum](https://w4.stern.nyu.edu/facdir/lpederse/papers/TimeSeriesMomentum.pdf) studies return persistence across liquid futures markets. It does not validate this particular RSI threshold, but it is a useful reminder that continuation is a serious competing hypothesis. A short-term rebound can occur inside a longer declining trend, while a longer recovery can follow an initially unsuccessful contrarian entry. It would be an error to invoke generic mean reversion as if every price series had a fixed equilibrium level to which it must quickly return. Earnings, inflation and risk-bearing capacity can shift the equilibrium itself.

For the November observation, the practical implication is to separate the state variable from the decision variable. RSI can identify that the recent price path is unusually one-sided. A decision still requires an investment horizon, an assessment of risk capacity, a benchmark and a view about the mechanism behind the move. Neither a value of 24.69 nor a historical average of 11.6% supplies those missing inputs. An investor already fully exposed to equities faces a different decision from one rebuilding an underweight position, even though both observe the same chart. The former considers whether incremental risk is justified; the latter considers whether a partial rebalance is preferable to continued cash exposure.

There is an additional timing distinction between detecting a signal and earning its measured return. A closing-price indicator becomes known only when the close is established. A backtest that enters at that same close must explain how an executable order could have been submitted without knowing the final input in advance. Trading at the next opening price avoids that particular look-ahead issue but introduces overnight price changes and opening execution costs. A market-on-close implementation may be feasible with an earlier estimate, yet that is a different rule with possible classification error around 30. This does not invalidate studying closing signals. It means that an indicator's descriptive accuracy and a strategy's achievable return must be evaluated separately, especially when the supposed edge over the unconditional benchmark is only a few percentage points.

 

Forty-one signals are not forty-one independent experiments

A backtest covering more than half a century sounds expansive, but the relevant statistical sample is the number of economically distinct opportunities, not the number of dates in the price file. If 41 observations contain clusters from the same recession or liquidation episode, they share important shocks. Two entries a few days apart can have almost identical three-month outcomes. Even signals separated by several months can have substantially overlapping 12-month holding periods. Counting each observation as an independent experiment makes the evidence look more precise than it is. The first task in auditing the exhibit is therefore to reconstruct the event calendar and identify which observations are distinct episodes rather than repeated measurements of the same shock.

The natural starting point is an explicit event rule. One candidate records every close below 30. Another records only a transition from at least 30 to below 30. A third allows a new signal only after a minimum waiting period. Each answers a different question. The daily rule weights prolonged oversold episodes heavily; the crossing rule can count repeated oscillations around the boundary; the waiting rule reduces clustering but embeds an additional design choice. No alternative should be selected simply because it reproduces the displayed total. Replication should follow the documented rule and then explain the 41-event count, not reverse-engineer a favorable definition from the desired answer.

Even under unrealistically favorable independence assumptions, the uncertainty is appreciable. As an illustration, a sample proportion of 0.80 based on 40 independent observations has a standard error near sqrt(0.80 × 0.20 / 40), or 6.3 percentage points. A rough two-standard-error range extends from approximately 67% to 93%. This is not a formal confidence interval for the chart: the exact completed sample sizes at three and six months are not provided, and independence is unverified. It is an order-of-magnitude demonstration. The headline 80% should not be treated as a known physical probability. Dependence, regime changes and selection of the rule can all add uncertainty beyond that simple calculation.

The 12-month figures make sample accounting more concrete. A positive share of 74.4% among 39 completed windows is consistent with approximately 29 positive outcomes and 10 nonpositive outcomes after rounding. One outcome switching classification changes the hit rate by roughly 2.6 percentage points. Moreover, 39 completed windows out of 41 total signals means that two events do not contribute to that particular statistic. A recent signal cannot yet have a completed one-year return. The appropriate response is to report horizon-specific denominators and exclude unavailable outcomes, not fill them with zeros, projected returns or shorter holding periods. Otherwise the comparison mixes completed experiments with assumptions about the future.

Right-censoring also affects how a reader interprets the newest observation. The November 20 signal is an application of a historical rule, not another completed confirmation of that rule's success. Its forward three-, six- and 12-month outcomes are still unknown at the time of the exhibit. A clean evaluation would distinguish the estimation sample from the current live observation and preserve that distinction when the chart is updated. It should also specify whether the most recent completed historical observations are drawn from the same data cutoff at every horizon. Without that bookkeeping, a smooth curve can conceal changing sample composition, particularly near the end of the dataset.

The mean-return comparison requires still more information. Standard errors for 4.2%, 7.6% and 11.6% cannot be recovered from their averages alone. One needs the dispersion of event returns, their dependence and the contribution of extreme outcomes. A few very large rebounds can raise the mean while leaving the median much less impressive. Reporting the median, interquartile range, lower-tail quantiles and the best and worst episodes would show whether the advantage is broad or concentrated. Comparing an event mean with an all-period mean also requires accounting for shared observations: the event windows are generally contained within the all-period sample, so those two estimated means are not independent datasets.

A sensible inferential approach would preserve time dependence rather than randomly shuffle individual returns. Resampling blocks of dates, or grouping entries into economically coherent episodes, can keep some of the clustering intact. Leave-one-episode-out analysis is especially informative with a small event count: remove each major cluster in turn and ask whether the sign and scale of the advantage survive. The objective is not to manufacture a single decisive significance statistic. It is to learn whether the result depends on one exceptional rebound or one particular market regime. If deleting one episode removes the estimated edge, the honest conclusion is that the hypothesis remains fragile.

Benchmark construction matters as well. “All periods” may mean every trading-day starting point, every calendar month or another schedule. Holding periods may use calendar dates or fixed trading-day counts. Price-only returns differ from total returns with dividends reinvested, especially over a full year. A historical S&P 500 price series is not identical to the realized returns of a modern exchange-traded fund with fees and distributions. None of these choices is necessarily wrong, but the selected and unconditional samples must use the same conventions. Otherwise part of the apparent advantage may come from inconsistent measurement rather than the oversold condition. The 1970-versus-1971 discrepancy makes documenting these basic choices particularly important.

Finally, a familiar threshold is not immune to data mining merely because it is widely used. Analysts can vary the indicator window, threshold, entry delay, exit horizon, reset rule and market universe. Reporting only the favorable combination understates the number of implicit experiments. A credible extension would freeze the current rule, test neighboring specifications transparently, reserve later observations for genuine out-of-sample evaluation and report failures alongside successes. For the present chart, the correct evidentiary posture is neither statistical dismissal nor unwarranted certainty. Forty-one reported signals justify investigating a plausible conditional pattern. They do not justify claiming a precisely known 80% success probability, a stable 2.9-percentage-point premium or a reliable ceiling on future losses.

 

Why recoveries can pay without being an arbitrage

A credible oversold argument needs an economic mechanism because a low indicator value, by itself, cannot cause cash flows to improve. One possibility is that prices fall faster than fundamentals because urgent sellers demand immediacy. Redemptions, margin calls and portfolio risk limits can force transactions when other investors are reluctant or unable to absorb inventory. A buyer who supplies balance-sheet capacity may receive a price concession. If that pressure later dissipates without a corresponding collapse in expected cash flows, a recovery can follow. In this interpretation, the apparent opportunity is compensation for taking the other side of an uncomfortable transfer, not a reward for recognizing a universally misunderstood chart pattern.

Research provides a framework for this mechanism without establishing the specific November trade. Stefan Nagel's [Evaporating Liquidity](https://www.nber.org/system/files/working_papers/w17653/w17653.pdf) interprets short-term reversal strategies as proxies for liquidity-provision returns and links their variation to market stress. Its relevance is that unusually attractive reversal returns can coexist with impaired liquidity supply. It is not a replication of a 14-day S&P 500 RSI strategy, and results from cross-sectional reversal portfolios cannot be transferred mechanically to an unhedged index position. The economic analogy is useful precisely because it explains why an apparent return premium might remain available: earning it can require deploying capital when carrying additional risk is especially difficult.

A second mechanism is a rise in required returns. In a simple valuation representation, price equals the discounted value of expected future cash flows. Holding the cash-flow path fixed, a higher required return reduces today's price and raises the return available to a new buyer if the investment survives and the assumed cash flows arrive. This does not require irrational selling. Investors may rationally demand more compensation when uncertainty rises, financing becomes scarce or their own wealth falls. A subsequent favorable average return can then reflect payment for bearing bad-state risk rather than an inefficiency. The distinction matters because a premium earned during stress cannot be assumed to provide protection against that same stress.

A numerical example separates the valuation channels. Imagine a perpetual cash flow of 5 units valued at a required return of 5%, giving a price of 100 under a deliberately simplified no-growth model. If the required return rises to 6% while the cash flow stays at 5, the price falls to about 83.3. If the required return later normalizes, there is room for a substantial rebound without any improvement in operations. But if sustainable cash flow simultaneously falls to 4, the price at 6% is about 66.7. Returning the discount rate to 5% then supports a price of only 80, not 100. These are illustrative calculations, not estimates of the S&P 500's current fair value. They demonstrate why distinguishing discount-rate news from cash-flow news is central to interpreting an oversold decline.

Behavioral amplification is another possible contributor, but it should not become a catch-all explanation. Investors may extrapolate recent losses, react to salient headlines or sell to reduce the discomfort of holding a visibly declining asset. Such behavior can reinforce price pressure temporarily. Yet observing a decline and then naming fear does not prove mispricing. The same sales may be rational responses to deteriorating information or binding constraints. To move beyond storytelling, an analyst would seek evidence that selling pressure is disproportionate to changes in expected cash flows, and that the pressure subsequently subsides. Measures of market breadth, trading conditions, fund flows and earnings revisions can help frame that question, though none is a complete or error-free identifier.

These mechanisms suggest distinct regimes with different implications. In a liquidity shock with resilient earnings expectations, a low RSI may flag a temporary concession that patient capital can exploit. In a prolonged fundamental contraction, it may simply record the first stage of a continuing repricing. In a discount-rate shock, the subsequent outcome depends partly on whether rates stabilize and on how much duration is embedded in equity valuations. A broad index can contain all three channels at once. Large growth companies, economically sensitive firms and defensive businesses may respond differently even while the index produces one headline indicator. This is why index-level RSI should not be used to infer that every constituent offers the same expected return or downside exposure.

The mechanisms can be made falsifiable without pretending to observe them perfectly. A liquidity-reversal hypothesis would be weakened if trading conditions normalize but prices fail to stabilize while earnings expectations continue to deteriorate. A discount-rate explanation would be incomplete if yields retreat substantially yet equity weakness broadens because cash-flow expectations fall faster. Conversely, stabilization in financing conditions alongside less negative revisions would support, but not prove, a recovery case. These are conditional diagnostics rather than automatic trading triggers. No contemporaneous readings for those variables are supplied by the RSI exhibit, so assigning the November event to one regime would go beyond the evidence. The chart tells us about past conditional outcomes; it does not diagnose today's causal mix.

Risk adjustment follows naturally from this framework. Suppose the events with the highest future returns are also those with the worst interim volatility or the greatest sensitivity to recession. A higher arithmetic return then need not imply a higher risk-adjusted return for an investor who already holds substantial equity exposure. The additional position can increase losses in precisely the states where the rest of the portfolio is under pressure. Assessing the signal requires examining conditional volatility, correlations, tail losses and capital usage, not only the difference between 7.6% and 4.7%. The relevant question is whether the incremental expected reward compensates for the incremental risk at the portfolio level, after uncertainty about that reward is acknowledged.

This economic interpretation also resolves the apparent paradox of a familiar signal retaining some usefulness. A broadly known observation need not be free money if exploiting it requires patience, financing capacity and willingness to tolerate losses before recovery. Nor does familiarity guarantee that the pattern will survive changing market structure. Faster execution, broader adoption and different investor constraints can alter who provides liquidity and at what price. The strongest conclusion is therefore modest but substantive: the chart's recovery pattern is compatible with established ideas about risk premia and liquidity provision, while remaining compatible with severe failures in particular episodes. That makes the signal a candidate input to a disciplined allocation process, not an arbitrage and not a substitute for analyzing fundamentals.

 

Terminal returns do not finance the journey

An investor earns a terminal return only by remaining invested long enough to reach the terminal date. That sounds obvious, but it is the central economic limitation of attractive forward-return charts. An institution with redemption obligations, an individual funding near-term expenses and a leveraged trader can all be forced to exit before a recovery arrives. The reported 14.3% maximum drawdown is therefore not a secondary detail beside the 11.6% average one-year return. It points to the possibility that the path can defeat the investor even when the eventual endpoint looks favorable. A realistic application must ask who supplies the capital, under what constraints and for how long that capital can remain committed.

Loss arithmetic explains why the path matters even without leverage. A position falling from 100 to 85.7 has lost 14.3%. Returning from 85.7 to 100 requires a gain of approximately 16.7%, calculated as 100 / 85.7 - 1. A later positive return measured from the original entry can still follow that drawdown, but the recovery requires a larger percentage move from the trough. If money is withdrawn during the decline, fewer units remain to participate. If risk limits force sales near the low, the investor's realized return differs sharply from the chart's buy-and-hold window. These calculations illustrate mechanics; they do not claim that every historical signal followed this particular sequence.

A simple exposure example translates market risk into portfolio risk. Suppose an otherwise uncorrelated cash reserve is used to establish an incremental equity allocation equal to 10% of portfolio value. A 14.3% decline in that allocation would reduce total portfolio value by about 1.43 percentage points before other positions move. At a 25% allocation, the same isolated shock would contribute roughly 3.58 percentage points. But these are partial-accounting examples, not portfolio forecasts. Existing equities, credit and other growth-sensitive holdings may decline at the same time. An investor already carrying substantial cyclical risk should not assess the new allocation as though it were the only position exposed to the shock.

Nor should the historical maximum become the stress scenario by default. A sample maximum is an observed realization, not a bound imposed by market structure. A future event can exceed it, and an uncertain drawdown definition makes direct calibration especially risky. A robust sizing exercise might ask whether the position remains tolerable under substantially larger hypothetical declines, prolonged recovery delays and unfavorable correlations. The chosen scenarios need not be presented as probabilities. Their purpose is to reveal which constraints fail first: liquidity needs, margin capacity, governance limits or the willingness to continue holding. A position that is survivable only if the historical maximum is never exceeded is not genuinely sized for uncertainty.

Leverage intensifies these issues. A simplified constant-notional position with twice the initial portfolio's exposure would translate a 14.3% underlying loss into roughly 28.6% of initial capital before financing costs and other effects. Real leveraged vehicles may rebalance daily, so their cumulative returns depend on the sequence of gains and losses rather than a simple multiple of the endpoint. Futures also require variation margin, which can create immediate cash demands during an otherwise temporary decline. An expected recovery six months later cannot pay tomorrow's margin call. The higher historical conditional mean does not eliminate this financing mismatch, and a small estimated advantage over the unconditional return does not automatically justify additional leverage.

The same path dependence undermines casual proposals to add a tight stop-loss to the historical strategy. The exhibit's forward returns assume some particular holding convention; adding a stop changes the return distribution and requires a new test. A stop may reduce some losses but also remove positions before the rebounds that drive the average. Gaps can produce execution below the intended threshold, and repeated exits and reentries can accumulate costs. Conversely, refusing every exit regardless of changing fundamentals is not disciplined merely because a chart shows positive averages. The point is methodological consistency: entry, exit, reentry and sizing rules must be evaluated together rather than assembling their attractive features from different untested strategies.

The opportunity-cost comparison must also match the investor's actual alternative. A fully invested benchmark already participates in market recoveries. A tactical overlay should therefore be judged on the incremental exposure and its funding source, not credited with the entire subsequent index return. An investor waiting in cash faces a different comparison involving cash yield, timing risk and foregone equity participation. An investor selling bonds to buy equities changes duration, liquidity and cross-asset sensitivity at the same time. The all-period return series is a useful statistical baseline, but it is not a complete counterfactual portfolio. Without an explicit alternative, even an accurately measured 7.6% average six-month return says little about the value added by the decision.

Implementation further narrows the usable edge. Bid-ask spreads, commissions where applicable, fund expenses, tracking differences, financing charges and taxes can separate index returns from realized results. For a liquid index instrument and infrequent trading, some costs may be modest, but they should be measured rather than assumed away. The more consequential issue can be cash drag between sparse signals. A strategy that spends most of its life uninvested may show attractive average event returns while underperforming a continuously invested benchmark over decades. To assess that strategy, one needs a full calendar-time portfolio series with a cash-return assumption and rules for overlapping signals. Event averages cannot supply a compound annual growth rate.

Finally, the headline averages do not identify an optimal position size. A theoretical sizing formula based on expected excess return divided by variance is highly sensitive to uncertainty in the mean, especially with a small and dependent event sample. Feeding the full historical advantage into such a formula risks treating estimation noise as investable conviction. A more defensible use of the chart is to modestly inform a pre-existing risk budget, with exposure limited by the investor's ability to withstand adverse paths. The distinction is consequential: a signal can improve the timing of a planned rebalance without becoming the foundation of a leveraged directional bet. Terminal return evidence has value only when the portfolio can finance the journey required to earn it.

 

A disciplined way to use the November signal

The November reading is most useful when it changes the quality of a decision rather than replacing the decision process. An investor can acknowledge that the historical evidence leans toward recovery over several months while refusing to infer an immediate floor. That stance is not indecision. It distinguishes a probabilistic input from an instruction that ignores portfolio context. The reported gap between signal and unconditional returns is large enough to merit attention but uncertain enough to demand restraint. A practical framework should therefore connect the indicator to an existing allocation objective, specify the capital at risk and identify what additional information would strengthen or weaken the case.

Consider an illustrative investor whose strategic equity target is 60% and whose current exposure has fallen to 55% after a market decline. Rebalancing toward the target is already supported by the investment policy; the oversold observation may inform the timing or staging of that adjustment. Restoring one or two percentage points of exposure now and reassessing later is economically different from raising the target itself to 70% because the chart looks favorable. The first action reduces a deviation from policy, while the second introduces a new tactical risk budget. Neither is automatically correct, but the distinction prevents an indicator about recent price pressure from quietly becoming a wholesale change in long-term risk preference.

Staging an allocation also has a real tradeoff rather than an unqualified benefit. If the market rebounds immediately, delayed purchases will participate less and can underperform an immediate rebalance. If prices continue falling, retaining some liquidity reduces the initial exposure and preserves flexibility. The choice depends on the cost of being early relative to the cost of missing a rebound, not on an ability to avoid both outcomes. A precommitted schedule or a clearly specified reassessment rule can reduce emotional reactions to subsequent price moves. Repeatedly changing the schedule after each headline would undermine that benefit and create an unrecorded discretionary strategy that cannot be evaluated against the original plan.

An investor already at the strategic target should ask a more demanding question before adding exposure. Does the expected incremental return, after shrinking a noisy historical estimate toward a more conservative prior, justify the additional downside concentration? There is no reliable numerical answer in the exhibit because the required risk distribution is missing. The reported 80% hit rate is not enough to calculate expected utility, and the 14.3% drawdown is not enough to describe the tail. In that setting, doing nothing can be a coherent implementation of the evidence. A signal need not trigger a trade to be useful; it can discourage panic selling or sharpen the conditions under which a future rebalance would occur.

Monitoring should focus on whether the proposed mechanism remains plausible. A liquidity-recovery thesis calls for attention to trading conditions and funding availability. A valuation thesis requires attention to earnings expectations and the discount rate. A portfolio thesis requires attention to cash needs and aggregate risk exposure. These are different questions and should not be collapsed into a dashboard where every improving line is counted as independent confirmation. Many indicators respond to the same underlying shock and can double-count information. Nor should a higher RSI automatically be interpreted as proof that fundamentals have improved. It may simply reflect a less negative recent price path, which is valuable information but not the same information.

The research process can be strengthened with a prospective record. Freeze the indicator calculation, event rule, entry timing and holding horizons before observing the next outcomes. Record the November 20 observation as a live event and maintain separate status fields for incomplete three-, six- and 12-month windows. When an outcome matures, compare it with both the original conditional distribution and a consistently constructed benchmark. A single success would not validate the strategy, just as a single failure would not disprove a probabilistic claim. What matters is whether a series of genuinely subsequent observations remains compatible with the hypothesized advantage after costs and with realistic portfolio constraints.

Replication should proceed in a clear order. First reconcile the 1970 and 1971 start-date descriptions and obtain the full event list. Then reproduce the latest 24.69 reading and the March 20 value of 29.88 using the stated smoothing convention. Next reproduce event counts, horizon-specific completed samples, return averages, hit rates and the drawdown definition. Only after those steps should an analyst test alternative thresholds, execution delays or filters. This ordering prevents new specifications from obscuring an inability to reproduce the original claim. It also allows disagreement to be localized: a mismatch in raw prices is different from a mismatch in signal rules, and both are different from an economically weak result after transaction costs.

There is a useful hierarchy of evidence here. The observed chart establishes what the displayed study reports. A reproducible dataset would establish that the calculations follow from explicit inputs. A robust statistical design would establish how much confidence to place in the estimated conditional relationship. A realistic portfolio simulation would establish whether a defined investor could have captured it. Subsequent live observations would then test whether the relationship persists. Skipping from the first layer to the last invites overconfidence. Conversely, demanding certainty before allowing a small, policy-consistent rebalance would set an impossible standard for most investment decisions. The appropriate evidentiary burden should rise with the size, leverage and irreversibility of the contemplated action.

The bottom line returns to the chart's full message rather than its most attractive annotation. The S&P 500's reported 24.69 RSI identifies unusually one-sided recent weakness. Historical signal windows show average returns of 4.2%, 7.6% and 11.6% at three, six and 12 months, against 2.3%, 4.7% and 9.6% across all periods. Yet the apparent advantage is not uniform across horizons, the 12-month hit rate is below the plotted unconditional rate, only 39 one-year windows are complete, and substantial drawdowns remain possible. Those facts support patient attention to a recovery opportunity, not a declaration that downside has ended. Used well, the November signal can improve rebalancing discipline and the formulation of testable hypotheses. Used carelessly, it converts a small historical sample into false certainty precisely when risk-bearing capacity matters most.

Comments


© 2035 by Someo Park Investment Management LLC.

bottom of page