Working paper. Not peer reviewed. Version of 22 November 2026.

Does the gold-to-equity ratio revert to its mean? A test of a valuation rule on United States data, 1973–2026

Gopala Subramanium
Independent researcher. Correspondence through the project repository.

Abstract

A common rule of thumb says that when gold is expensive relative to shares it should be trimmed, and when it is cheap it should be bought. I wrote that rule down in a precise form: a five-year rolling Z-score of the ratio of gold to an equity total-return index, a gold weight that slides between 5% and 15% of the portfolio, and two macro cross-checks. I then tested it on 644 months of United States data from open sources. The rule did not work. Acted on one month after each reading, the prescribed tilt returned −37 basis points a year relative to a constant 10% holding (Newey–West t = −2.60; p = 0.015 in a simulation that reproduces the monthly averaging in the data; stationary-bootstrap 95% interval −64 to −14). The loss is concentrated in the first half of the sample (−60 basis points a year, t = −2.41). In the second half the result is −15 basis points (t = −1.12), which cannot be told apart from zero. A high Z-score was followed by gold outperforming equities, the opposite of what the rule assumes, but that slope is not statistically reliable once overlapping forecast windows are allowed for (simulated p = 0.11). In no period did the rule earn a dependable profit. Two parts of the framework fare better: a total-return index matters a great deal for long-horizon comparisons, and a low equity earnings yield relative to the bond yield was followed by equities lagging bonds, mostly because bonds did well. I report the data, code and every test, including those that did not go the way I expected.

Keywords: gold, equities, valuation ratios, mean reversion, time-series momentum, tactical asset allocation. JEL: G11, G12, G17.

1. Introduction

Charts of gold divided by a share index are a fixture of market commentary. The usual reading is that a high ratio means gold is dear and a low one means it is cheap. The idea is intuitive, it resembles the logic of valuation ratios for equities (Campbell and Shiller 1998), and it has an obvious appeal for anyone deciding how much gold to hold.

I set out to build a careful version of that idea. The naive ratio has known defects: it uses a price index and so ignores dividends, in some countries it uses a local gold price distorted by import duties, and it compares today with a fixed long-run average although the level of the ratio has shifted across monetary regimes. The version I specified corrects each of these. It uses a total-return index, a gold price rebuilt from the dollar spot price and the exchange rate, and a rolling five-year window. It converts the result into a Z-score and maps the Z-score to a target gold weight.

The rule's parameters (the window, the tilt, the band and the thresholds) were set in the project's original specification before any backtest was run, and none was tuned afterwards. That specification was not registered publicly, so the reader has my word for the order of events and nothing more. A rule tuned on the data it is tested on will look good by construction; a rule fixed in advance can fail. The design of the test is a different matter. The start of the sample, the horizons, the lags and the ranges in Table 2 were not registered anywhere in advance. I chose them while doing the work, and I report how sensitive the conclusions are to them.

The literature on gold is large (O'Connor, Lucey, Batten and Baur 2015 survey it), and several of its themes bear on this question: Erb and Harvey (2013) examine whether the real price of gold reverts, Baur and Lucey (2010) and Baur and McDermott (2010) study gold as a hedge and safe haven, and Barsky and Summers (1988) link gold to real interest rates. I have not found a published test of this particular rolling-window ratio rule, with its weights and gates, on a half-century sample with open data and code.

2. The rule under test

Let G be the dollar gold price per troy ounce, FX the local-currency price of a dollar, and TRI an equity total-return index. For month t:

Clean goldt = Gt × FXt ÷ 31.1034768 (local currency per gram)

Rt = Clean goldt ÷ TRIt

Zt = (Rt − μt) ÷ σt

where μt and σt are the mean and sample standard deviation of R over the sixty months ending at t. The target gold weight is

Wt = max(5%, min(15%, 10% − 2.5% × Zt)).

Two gates sit beside it. Gate 1 reads the ten-year real yield and treats a level above 2.25% as a headwind for gold. Gate 2 computes the equity earnings yield minus the ten-year government bond yield and treats a spread below −3 percentage points as a sign that equities are expensive. A seven-branch decision table combines Z and the gates into instructions such as "trim gold to the floor" or "accumulate". Z is unaffected by the units of G, FX or TRI.

The rule embeds three testable claims. H1: a high Z is followed by gold underperforming equities. H2: following W earns more than holding 10% throughout. H3: the gates identify periods that are worse for gold (Gate 1) or for equities relative to bonds (Gate 2).

3. Data

Gold is the World Bank's monthly average price in dollars per troy ounce. Equities are Robert Shiller's S&P Composite series: price, dividends, earnings, the consumer price index and the ten-year Treasury yield (Shiller 2015). Both were taken from public GitHub repositories that republish them under the Open Data Commons Public Domain Dedication and License, and the equity file is extended there with monthly-average prices from FRED. The sample runs from January 1968 to September 2026. I start in 1968 because the two-tier gold market set up in March of that year created a free-market price separate from the official one. The first sixty-month Z-score is December 1972.

Total return is built in the standard way from Shiller's data: each month's return is (Pt + Dt/12) ÷ Pt−1. Four limits:

  • Prices on both sides are monthly averages of daily values, not month-end closes. A reading for month t is known only when the month is over, but the change from one monthly average to the next starts to accrue half-way through month t, and averaging itself induces positive autocorrelation in measured returns (Working 1960). To keep the two apart, every test acts one full month after the reading: the weight from month t is first applied to the return from month t+1 to t+2, and forecast windows start at month t+1. This leaves 644 months of returns.
  • Until August 1971 the free-market gold price moved in a narrow range, so the earliest five-year windows straddle the end of dollar convertibility. Section 5.5 repeats the test with the data starting in September 1971 and in January 1975. The source records the monthly average to the nearest dollar throughout.
  • The republished copy of Shiller's dividend, earnings and yield series used here stops in June 2023, although Shiller's own file runs later. From then on the price index comes from FRED and I carry the last dividend forward at 5% a year. Dividends add about 1.3% a year to total return in those months, so the assumption has little room to matter, but it is an assumption. Tests of the gates stop where the earnings and yield data stop. Shiller's earnings and price-index figures are interpolated and are published with a delay, so the gate tests use information slightly earlier than an investor would have had it.
  • Inflation-indexed Treasury securities were first issued in 1997 and a ten-year constant-maturity yield is published only from 2003. For Gate 1 I use the ten-year yield minus trailing twelve-month inflation, which is a backward-looking stand-in for a real yield.

No test result in this paper was entered by hand. One script assembles the dataset from the two source files and a second runs every test.

4. Method

Predictive regressions. For horizons k of 1, 12, 36 and 60 months I regress the log return of gold minus the log return of equities, over the k months that begin one month after the reading, on Zt. Overlapping horizons make the errors serially correlated, so standard errors are Newey and West (1987) with k lags. With long overlaps these t-statistics reject too often (Hodrick 1992), and a persistent regressor biases the slope as well (Stambaugh 1999), so I add two checks: the same regression on one reading a year, which has no overlap, and simulated p-values, described below.

Out-of-sample fit. Following Campbell and Thompson (2008) and Welch and Goyal (2008), I compare twelve-month forecasts from an expanding-window regression with forecasts from the running mean, using at each date only outcomes already realised. Forecasts begin in July 1990.

The tilt. The return to following the rule, relative to a constant 10%, is (Wt − 10%) × (gold return − equity return) over the month from t+1 to t+2. I report its annualised mean, a Newey–West t-statistic with twelve lags, and a stationary-bootstrap interval (Politis and Romano 1994) with a mean block of 24 months and 5,000 resamples. Portfolio statistics include a cost of 0.25% of the amount of gold weight changed.

Simulated p-values. I simulate 1000 pairs of daily random walks with the same volatility, drift and correlation as gold and equities, average them by month as the data are averaged, and run the same tests on each pair. By construction the rule has no edge in these series. The t-statistic for the tilt behaves as it should there, exceeding 1.96 in 5.7% of runs. The t-statistic for the twelve-month regression does not: it exceeds 1.96 in 15.6% of runs, and its 5% critical value is 2.68. The p-values I quote are the share of simulated runs at least as extreme as the data.

Robustness. I repeat the tilt with windows from three to twenty years and an expanding window, with stronger tilts and wider bands, with later starting dates, and in each half of the sample. Trying many variants invites false discoveries (White 2000; Harvey, Liu and Zhu 2016); here the variants are used to see whether a failure is robust, which is the safe direction.

Persistence. I test the log ratio for a unit root (Dickey and Fuller 1979) and for stationarity (Kwiatkowski, Phillips, Schmidt and Shin 1992), and estimate a first-order autoregression.

5. Results

5.1 Any pull back to a level is slow

Over the full sample the augmented Dickey–Fuller test does not reject a unit root in the log ratio (statistic −1.19, p = 0.68 with a constant; p = 0.38 with a trend), and the KPSS test rejects stationarity (statistic 2.63, p at or below 0.01). The verdict depends on where the sample starts: from 1980, close to gold's peak, the Dickey–Fuller test does reject (statistic −3.23, p = 0.018). The autoregressive coefficient is 0.9989 with a standard error of 0.0022, too close to one to say how fast the ratio reverts, or whether it does. What can be said is that any reversion is slow next to a five-year window.

The Z-score does not behave like a standard normal variable. Its standard deviation is 1.54, it ranged from −2.49 to +4.90, and it was beyond ±1.5 in 15% of months on the high side and 15% on the low side, where a normal variable would be there about 7% of the time each way. It was inside the "neutral" band of ±0.5 in 13% of months, against 38% for a normal variable. The rule's thresholds were set as if Z were normal.

5.2 H1: a high Z was not followed by gold lagging

Table 1. Log return of gold minus equities over the k months starting one month after the reading, regressed on Z. Newey–West t-statistics. "First half" ends before November 1999.
Horizon (months)Full sampleFirst halfSecond halfR² (full)
SlopetSlopetSlopet
1+0.0051+2.27+0.0078+2.00+0.0016+0.600.017
12+0.0498+2.16+0.0637+1.54+0.0219+1.050.077
36+0.0778+1.55+0.0638+1.32+0.0562+0.960.055
60+0.1179+1.53+0.1202+1.14+0.0506+0.580.069

H1 requires negative slopes. Every estimate is positive. That is evidence against the rule, but it stops short of showing the opposite. At twelve months the t-statistic is 2.16, below the simulated 5% critical value of 2.68 (simulated p = 0.11). With one reading a year there are 53 observations and no overlap; the slope is positive for all 12 possible starting months, with t-statistics from 1.42 to 2.72 and a mean of 2.08. The relationship is weaker in the second half of the sample than in the first.

Out of sample, the regression forecast beat the running mean at twelve months over the whole forecast period (R² of 0.113). That is a result of the 1990s. From November 1999 the figure is 0.025, and from December 2013 it is −0.28: worse than the running mean.

5.3 What followed each kind of reading

Table 2. Gold return minus equity return over the following twelve months, by Z-score at the start (percentage points).
Z at the startMonthsMedianMeanGold ahead
below −1.598−19.5−18.425%
−1.5 to −0.5249−9.4−8.332%
−0.5 to +0.586+15.2+6.655%
+0.5 to +1.5110+3.9+12.163%
above +1.590+0.7+4.851%

When the rule called gold very cheap (Z below −1.5), gold went on to lag equities in roughly three cases out of four. When it called gold expensive (Z above +1.5), gold was ahead about half the time. The pattern is closer to a step than a slope: after readings below −0.5 gold usually lagged, and after readings above that it was level or ahead. These are overlapping windows from two long gold cycles and should be read as description.

5.4 H2: the tilt did not pay

Following the prescribed weight, one month after each reading, returned −37.2 basis points a year relative to a constant 10% in gold (t = −2.60, simulated p = 0.015). The bootstrap interval is −63.5 to −13.6 basis points, and 4 of 5000 resamples were positive. Acting in the same month as the reading gives −44.1 (t = −3.10), which flatters the case against the rule because of the averaging overlap; waiting two months gives −37.0 (t = −2.68). Turnover is low, about 3.9% of the portfolio a year, so costs are not the explanation. One reading of the full seven-branch decision table, tested where the gate data allow, returned −44.6 basis points (t = −2.67).

The loss is not spread evenly. In the first half of the sample the tilt returned −60.1 basis points a year (t = −2.41). In the second half, from November 1999, it returned −14.5 (t = −1.12; bootstrap interval −41.1 to +8.3), and from October 2008 it returned −10.9 (t = −0.69). Over the last quarter-century the rule made no detectable difference.

Table 3. The tilt by decade. "Gold minus equities" is the average annual return difference; "Tilt" is the annual return to the rule relative to a constant 10%, in basis points.
PeriodGold minus equities (% a year)Average ZTilt (bp a year)
1973–1979+27.3+1.29−96.2
1980–1989−15.3−0.53−23.1
1990–1999−20.3−1.49−76.8
2000–2009+14.5+1.07−29.6
2010–2019−9.7−0.54−21.8
2020–2026+1.3−0.18+26.7

The table shows the mechanism. In the decades when gold was winning, the average Z was high and the rule was underweight. In the decades when gold was losing, the average Z was low and the rule was overweight. The rule was on the wrong side in 5 of 6 periods. The exception is the most recent one.

Table 4. Portfolios from December 1972 to September 2026, after trading costs.
PortfolioReturn (% a year)Volatility (%)Return ÷ volatilityWorst drawdown (%)
Equity only11.0212.610.87−49.0
Gold only8.1317.190.47−62.1
Static 10% gold10.9511.480.95−43.9
Rule (5–15%)10.5311.460.92−46.4
Rule reversed (5–15%)11.3411.550.98−41.3

A constant 10% in gold went with lower volatility and a smaller drawdown than equities alone, at almost no cost in return, which is the conventional case for a strategic gold holding. Tilting that holding by this rule gave back about 0.4 percentage points a year.

5.5 Robustness

Table 5. The same test at other window lengths.
WindowTilt (bp a year)t12-month slopet
3 years−34.9−2.62+0.060+2.68
5 years−37.2−2.60+0.050+2.16
7 years−22.2−1.80+0.033+1.38
10 years−23.1−1.90+0.035+1.28
15 years−9.6−1.14+0.003+0.08
20 years−4.5−0.56−0.026−0.56
Expanding−18.9−1.39+0.024+0.55

The tilt is negative at every one of the 7 window lengths, and reliably so at 2 of them, the shortest. The loss shrinks as the window lengthens, which is what one would expect if the ratio moves in swings lasting a decade or more: a longer window is slower to bet against them. A stronger tilt or a wider band increased the loss.

Starting the data later weakens the result without changing its sign. With the series beginning in September 1971, after the dollar's link to gold ended, the tilt returned −30.1 basis points a year (t = −2.52). Beginning in January 1975 it returned −23.7 (t = −2.07). The years 1973 and 1974, when gold rose sharply while the rule was underweight, account for a large part of the full-sample loss.

5.6 H3: the gates

Gate 2. The earnings yield fell more than three points below the bond yield in 29 months (4.4% of the sample), grouped in 8 separate spells. Over the following twelve months equities returned 8.8 percentage points less than ten-year bonds on average, against 5.6 points more in other months. A regression of that return difference on the spread has a slope t-statistic of 3.54 and an R² of 0.10. Splitting the two legs shows where this comes from: the spread predicts bond returns (t = −4.69) and not equity returns (t = 0.87). When bond yields were high relative to earnings yields, bonds went on to do well. The gate pointed the right way for the choice between equities and bonds, on a small number of independent episodes, and comparisons of earnings yields with nominal bond yields have well-known conceptual problems (Asness 2003).

The branch that combines the two conditions never occurred. Of the 88 months with Z above +1.5, 0 also had the spread below −3.

Gate 1. The backward-looking real yield was above 2.25% in 44% of months. When Z was below −0.5, gold's next twelve-month return averaged −1.2% with the real yield above the threshold and 6.4% otherwise. The direction agrees with Barsky and Summers (1988), who tie the price of gold to real interest rates. A regression of gold's return on the real yield is not significant (t = −0.87). Both figures are below gold's average of 8.9% over the same months, which is H1 failing again: gold that screened as cheap did poorly either way.

5.7 Dividends

Over the sample the total-return index grew 4.9 times as much as the price index, a difference of 2.8% a year. Measured against the price index, the gold ratio ended at 1.49 times its starting level. Measured against total return, it ended at 0.30 times. Anyone comparing today's gold-to-index ratio with its level decades ago and ignoring dividends is making an error of that size.

Inside a five-year window the correction matters much less. The Z-scores computed with and without dividends have a correlation of 0.995 and differ by 0.14 on average. The total-return adjustment is right in principle and important for levels; it does not rescue the rolling rule.

5.8 A comparison: trend

If the ratio moves in long swings, a rule that follows the swing should have done better than one that bets against it. As a check I used a signal adapted from the time-series momentum literature and not fitted here: the sign of the past twelve months' return of gold relative to equities (Moskowitz, Ooi and Pedersen 2012 use each instrument's own excess return), tilting gold to 15% when gold had led and 5% when it had lagged, again acting a month late. It returned +49.8 basis points a year (t = 2.64). Almost all of that came in the first half of the sample (+87.2, t = 2.93); in the second half it returned +12.6 (t = 0.60). I report this to explain the main result. I looked at it after seeing the rule fail, and whatever edge it had is no longer visible.

5.9 Other markets

The same script was run, unchanged, on five further series built from published total-return indices, with gold converted at Federal Reserve monthly average exchange rates. These samples are much shorter than the main one, and the rule’s parameters were not adjusted for any of them. Of the five, three show a negative tilt, and none is significant at the 5% level on the simulated p-value.

Table 6. The rule on other series. Tilt is in basis points a year against a constant 10% in gold, acting one month after each reading.
SeriesTilt tested fromMonthsTilttSimulated p12-month slopeSlope p
India: MSCI India net total returnNovember 2005250+12.8+0.740.483−0.0770.207
United States (official index): S&P 500 Total ReturnDecember 1992405−28.0−2.050.086+0.0530.085
World: MSCI ACWI net total returnNovember 2005250−4.5−0.290.786+0.0100.805
Euro area: MSCI EMU net total returnNovember 2005250+1.5+0.100.920−0.0270.519
United Kingdom: MSCI United Kingdom net total returnNovember 2005250−6.3−0.410.727−0.0110.803

These series overlap heavily with one another and with the last part of the United States sample, so they are not independent confirmations. They are reported so that a reader can see the rule was not tested on one market alone.

6. Discussion

Why does a rule that sounds sensible fail? The ratio of gold to equities is the ratio of two assets with different long-run drivers. Equities compound with earnings. Gold has no cash flow, and its price in terms of equities has moved in long swings associated with inflation, real interest rates and confidence in the monetary system. Those swings lasted ten to twenty years. A five-year window treats the early part of each swing as an extreme and bets against it, then recalibrates just as the swing matures.

Two published results bear on this. Erb and Harvey (2013) argue that the real price of gold may revert to its mean, but over horizons far longer than any tactical window, and they stress how uncertain that is. Time-series momentum at horizons up to a year is documented across asset classes, including commodities and equity indices (Moskowitz, Ooi and Pedersen 2012). The pattern here fits between the two: persistence at short horizons before 2000, and no detectable reversion at the horizons the rule cares about.

The two halves of the sample tell slightly different stories, and both matter. Before 2000 the rule lost money to a degree that is hard to put down to chance. Since then it has done nothing measurable in either direction. Neither period offers any support for using it.

The failure does not make the ratio useless as a description. Knowing that gold buys many more index units than it did five years ago is a fact about recent history. What the evidence does not support is reading that fact as a forecast.

7. Limits

  • One main sample. The main test is on United States data. Section 5.9 reports the same test on five shorter series from other index providers. Those samples begin between 1988 and 2000, which is too short to settle much on their own.
  • Few cycles. The sample contains three large gold bull markets (the 1970s, 2001 to 2011, and the years since 2022) and one long bear market. Fifty years is a long time, but it is a small number of independent regimes.
  • Monthly averages. Month-end data would be cleaner. Acting a month late removes the overlap this creates, and the simulated p-values are built on series averaged in the same way.
  • The early years. Gold was barely floating before late 1971, and 1973–74 contributes a large share of the loss. Later starting dates give smaller, still negative, figures.
  • Choices made along the way. The rule's parameters were set before the test; the test design was not. The sensitivity tables are there so that a reader can see what depends on those choices.
  • Gate data. The real-yield series is backward-looking, trailing earnings can be distorted in recessions, and the bond return is approximated from yield changes.
  • What would change my mind. A negative and reliable slope in Table 1 on month-end data, or a positive tilt in several other markets, would reopen the question.

8. Conclusion

I wrote down a gold-to-equity valuation rule, tested it on more than fifty years of open data, and found no evidence that it works. Over the full sample the prescribed tilt lost about 0.37% a year against holding a constant 10%, with the loss concentrated before 2000. Since then it has made no measurable difference. Readings the rule calls cheap were more often followed by further underperformance than by recovery, though the evidence for that reversal is suggestive, not conclusive. The total-return correction is important for comparing levels across decades, and the equity-spread gate carried some information about equities against bonds. In this sample the tactical rule added nothing to a constant allocation.

Data and code

Everything needed to reproduce this paper is in the project repository at github.com/gopalasubramanium/aurumratio: scripts/prepare_us_longrun.py builds the dataset, scripts/study.py runs the tests and writes data/results/us_longrun.json, and the build step copies those numbers into this text. The calculator on the main page recomputes the regression, the table of outcomes and the tilt in the browser, with independent code, and agrees with the figures above.

This paper is provided for academic and educational purposes only. It is not investment advice. Past relationships do not guarantee future results.

References

  • Asness, C. S. (2003). Fight the Fed model. Journal of Portfolio Management, 30(1), 11–24.
  • Barsky, R. B., and Summers, L. H. (1988). Gibson's paradox and the gold standard. Journal of Political Economy, 96(3), 528–550.
  • Baur, D. G., and Lucey, B. M. (2010). Is gold a hedge or a safe haven? An analysis of stocks, bonds and gold. The Financial Review, 45(2), 217–229.
  • Baur, D. G., and McDermott, T. K. (2010). Is gold a safe haven? International evidence. Journal of Banking & Finance, 34(8), 1886–1898.
  • Campbell, J. Y., and Shiller, R. J. (1998). Valuation ratios and the long-run stock market outlook. Journal of Portfolio Management, 24(2), 11–26.
  • Campbell, J. Y., and Thompson, S. B. (2008). Predicting excess stock returns out of sample: can anything beat the historical average? Review of Financial Studies, 21(4), 1509–1531.
  • Dickey, D. A., and Fuller, W. A. (1979). Distribution of the estimators for autoregressive time series with a unit root. Journal of the American Statistical Association, 74(366), 427–431.
  • Erb, C. B., and Harvey, C. R. (2013). The golden dilemma. Financial Analysts Journal, 69(4), 10–42.
  • Harvey, C. R., Liu, Y., and Zhu, H. (2016). … and the cross-section of expected returns. Review of Financial Studies, 29(1), 5–68.
  • Hodrick, R. J. (1992). Dividend yields and expected stock returns: Alternative procedures for inference and measurement. Review of Financial Studies, 5(3), 357–386.
  • Kwiatkowski, D., Phillips, P. C. B., Schmidt, P., and Shin, Y. (1992). Testing the null hypothesis of stationarity against the alternative of a unit root: How sure are we that economic time series have a unit root? Journal of Econometrics, 54(1–3), 159–178.
  • Moskowitz, T. J., Ooi, Y. H., and Pedersen, L. H. (2012). Time series momentum. Journal of Financial Economics, 104(2), 228–250.
  • Newey, W. K., and West, K. D. (1987). A simple, positive semi-definite, heteroskedasticity and autocorrelation consistent covariance matrix. Econometrica, 55(3), 703–708.
  • O'Connor, F. A., Lucey, B. M., Batten, J. A., and Baur, D. G. (2015). The financial economics of gold: a survey. International Review of Financial Analysis, 41, 186–205.
  • Politis, D. N., and Romano, J. P. (1994). The stationary bootstrap. Journal of the American Statistical Association, 89(428), 1303–1313.
  • Shiller, R. J. (2015). Irrational Exuberance (3rd ed.). Princeton University Press. Data: econ.yale.edu/~shiller/data.htm.
  • Stambaugh, R. F. (1999). Predictive regressions. Journal of Financial Economics, 54(3), 375–421.
  • Welch, I., and Goyal, A. (2008). A comprehensive look at the empirical performance of equity premium prediction. Review of Financial Studies, 21(4), 1455–1508.
  • White, H. (2000). A reality check for data snooping. Econometrica, 68(5), 1097–1126.
  • Working, H. (1960). Note on the correlation of first differences of averages in a random chain. Econometrica, 28(4), 916–918.
  • World Bank. Commodity Price Data (The Pink Sheet). worldbank.org/en/research/commodity-markets.

Conceptualized & Published for Public Good by Gopala Subramanium.

Disclaimer — Academic & Educational Purposes Only: This tool, methodology, and all accompanying calculations are provided strictly for academic research, financial literacy, and educational purposes. Nothing on this page constitutes investment, financial, tax, legal, or portfolio management advice, nor an offer, solicitation, or recommendation to buy, sell, or hold any security, ETF, commodity, or asset class in India or any other jurisdiction. Historical relationships, statistical Z-scores, and macro filters do not guarantee future performance. Users are solely responsible for verifying all data inputs and must consult a qualified, regulated financial adviser in their respective jurisdiction before making any investment decision.