pith. sign in

arxiv: 2604.12062 · v2 · submitted 2026-04-13 · 📊 stat.ME

Is There an AI Bubble? Robust Date-Stamping for Periods of Exuberance

Abir Sarkar , Martin T. Wells This is my paper

Pith reviewed 2026-05-12 00:56 UTC · model grok-4.3

classification 📊 stat.ME
keywords bubble detectiondate-stampingstochastic volatilityADF testAI equitiesexuberancemoderate deviationsfinancial time series
0
0 comments X p. Extension

The pith

A volatility-robust ADF test date-stamps bubble start and end points even with persistent volatility changes.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper develops a practical way to detect and date the origination and collapse of speculative bubbles when prices follow autoregressive processes with highly persistent volatility. Standard right-tailed unit root tests can produce unreliable signals during volatility spikes, which are common in technology markets. By extending the Dickey-Fuller framework with moderate-deviation asymptotics, the new procedure supplies nuisance-parameter-free critical values that differ for bubble formation and burst. Monte Carlo results show improved power and fewer transient false positives compared with homoskedastic methods. An application to AI-exposed stocks finds pervasive exuberance that varies in timing, intensity, and length across firms such as Alphabet, TSMC, Tesla, and Nvidia.

Core claim

The paper claims that extending right-tailed Dickey-Fuller tests to autoregressive models with persistent mean and volatility dynamics, via moderate-deviation asymptotic theory, produces a stochastic-volatility-robust ADF test. This test delivers distinct critical values for origination and collapse without nuisance parameters, yields a consistent date-stamping estimator, and generates more stable alarms with reduced false positives around volatility spikes. Simulations document strong power gains over homoskedastic procedures when volatility persistence is pronounced. The empirical analysis of AI-related equities shows widespread bubble episodes with substantial heterogeneity in their dates

What carries the argument

The SV-ADF test, obtained by extending right-tailed Dickey-Fuller unit root tests to models with persistent mean and volatility dynamics and applying moderate-deviation asymptotics to obtain separate critical values for origination and collapse.

If this is right

  • The date-stamping estimator remains consistent and asymptotically tractable under the stated conditions.
  • The test produces fewer transient false positives around volatility spikes than standard homoskedastic procedures.
  • Application to AI-exposed equities reveals pervasive exuberance with clear differences in timing and duration across stocks.
  • Alphabet and TSMC exhibit especially strong bubble dynamics in the current cycle, while Tesla and Nvidia showed earlier pronounced episodes.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same procedure could be applied to other high-volatility sectors to compare bubble prevalence across asset classes.
  • Regulators might adopt the distinct origination and collapse thresholds to trigger earlier or later interventions than current methods allow.
  • Prior bubble studies that ignored persistent volatility may have reported overstated or understated episode lengths in tech markets.
  • Extending the framework to joint testing of multiple stocks could reveal common factors driving the observed heterogeneity.

Load-bearing premise

The moderate-deviation asymptotic theory applies to autoregressive models with highly persistent mean and volatility dynamics without requiring strict parametric structure on the volatility process.

What would settle it

Monte Carlo experiments in which the SV-ADF procedure exhibits size distortions or loses consistency when volatility follows paths outside the moderate-deviation regime.

Figures

Figures reproduced from arXiv: 2604.12062 by Abir Sarkar, Martin T. Wells.

Figure 1
Figure 1. Figure 1: The Nasdaq index together with the stock prices of Tesla, Broadcom, and Nvidia over the past ten years. The pronounced run-ups in these stocks during 2020–2025 are suggestive of speculative bubble dynamics. Data source: Yahoo Finance (Yahoo! Finance, 2026) [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Stock price for Broadcom and Nvidia over 2022-2026 exhibit pronounced heteroskedasticity. Volatility is calculated using a rolling window of the previous 8 weeks. The visual evidence supports a specification that allows for persistent variation in scale. The evidence in [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Proposed SV-ADF procedure identified speculative explosive bubble behaviour in individual stocks. The dotted red vertical lines mark bubble origination, the dotted green lines signify collapse. Speculative intervals: Nvidia (Jan’ 2024 – Aug’ 2024), Tesla (June 2020–March 2021). First, we develop a novel volatility-robust stochastic ADF test and its recursive sup-type implementation for retrospective bubble… view at source ↗
Figure 4
Figure 4. Figure 4: Misdating of Bitcoin and Nvidia bubble episodes under the PWY approach. This discrepancy likely reflects the reliance of homoskedastic thresholds in an environment where volatility is strongly time-varying, as is typical in cryptocurrency markets. SV-ADF procedure delivers date-stamps that are more economically plausible, aligned with major news developments. Finally, we provide an inferential theory to co… view at source ↗
Figure 5
Figure 5. Figure 5: Daily adjusted close prices of the Magnificent 7 and the six largest semiconductor/AI-infrastructure firms by market capitalization over the past 10 years. Panels are arranged in descending market-cap order, with the first six showing tech firms and the next six for the rest. The vertical line at January 2020 marks the onset of the surge emphasized in our analysis. Source: Yahoo Finance (Yahoo! Finance, 20… view at source ↗
Figure 6
Figure 6. Figure 6: Stock-price volatility for Nvidia over 2022–2025 exhibits pronounced peaks and troughs. PWY signals a spurious bubble window in one such high-volatility episode (June–November 2023). SV-ADF procedure identifies the most persistent and quantitatively pronounced episode of explosiveness (January–August 2024), rather than short-lived price spikes induced by volatility fluctuations. 4 Estimation and Inference … view at source ↗
Figure 7
Figure 7. Figure 7: Recursive SV-ADF statistic (blue) and daily stock price (red) for the six largest technology firms by market capitalization. Plotting both series provides visual inspection on whether threshold crossings in the recursive statistic coincide with meaningful price movements. The orange line denotes the 90% upper critical value under H0 used for bubble origination, while the brown line denotes the 10% lower cr… view at source ↗
Figure 8
Figure 8. Figure 8: Recursive SV-ADF Statistic (in blue) and daily stock price (in red) for 6 largest semiconductor/AI￾infrastructure firms by market capitalization. The sample period is 2022–2026, except ASML July 2018–2026. The color scheme is the same as [PITH_FULL_IMAGE:figures/full_fig_p018_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Date-stamping bubble behaviour in Nasdaq index. Time period for upper panel: 2020–2026, no exuberance detected. Time period for lower panel: 1982–2012, significant exuberance detected in 1995-2000. Date-stamps are accurate and provide early signals as compared to PWY. The color scheme is the same same as [PITH_FULL_IMAGE:figures/full_fig_p019_9.png] view at source ↗
Figure 10
Figure 10. Figure 10: Recursive SV-ADF test suggests significant bubble in Bitcoin and Ethereum during January 2021 – June 2021. Similar cryptocurrencies during this era exhibited pronounced exuberance (see the Appendix). The color scheme is the same as [PITH_FULL_IMAGE:figures/full_fig_p020_10.png] view at source ↗
Figure 11
Figure 11. Figure 11: Estimated autoregressive coefficients, growth rates, and their 95% C.I over 2022–2026. For most firms, unity lies within the interval for the AR coefficient, consistent with near-unit-root behavior. 5.1 Threshold Selection Insights In this section, we discuss the calibration of the cross-validated thresholds used in the SV-ADF tests and, in turn, the decision rules for accepting or rejecting the hypothesi… view at source ↗
Figure 12
Figure 12. Figure 12: One simulated time series path (in black) and recursive SV-ADF statistic (in blue). The red and brown vertical lines denote the true re = 0.3, rf = 0.6 respectively, while the blue and green mark the estimated rˆe = 0.308, rˆf = 0.652. The orange horizontal line is the origination threshold (log(ns)/10), purple line is collapse threshold (log(ns)/2). As explained in Section 4, the conventional PWY procedu… view at source ↗
Figure 13
Figure 13. Figure 13: The panels report the average value of Xτf −Xτe on a logarithmic scale under homoskedasticity (left) and stochastic volatility (right). In both panels, re = 0.4, rf = 0.6, c = 1, and α = 0.5. The sample size ranges from n = 50, 100, . . . , 500, while the number of Monte Carlo replications is fixed at B = 500. The visual evidence indicates that this difference grows exponentially with the sample size, whi… view at source ↗
Figure 14
Figure 14. Figure 14: The recursive SV-ADF test indicates a pronounced bubble episode in Cardano and Solana during 2021. By contrast, Binance Coin exhibited only weak and short-lived exuberance, while Dogecoin showed no evidence of a bubble. A.8. Threshold selections: cvβ,r and cvβ,l n cvδ 0.10,H0 cvδ 0.10,H1 500 0.7463 -0.2566 550 0.8563 -0.9916 600 0.8304 1.2746 650 0.0597 1.9418 700 0.8997 1.0183 750 0.7037 2.6310 800 0.982… view at source ↗
Figure 15
Figure 15. Figure 15: Threshold plots under H0 (red) and H1 (blue) in the settings as discussed in [PITH_FULL_IMAGE:figures/full_fig_p041_15.png] view at source ↗
read the original abstract

The recent surge in valuations among AI related firms has renewed concerns that markets may be entering a new phase of speculative exuberance, especially in the technology and semiconductor sectors at the center of the AI investment wave. This paper develops a practical econometric framework for detecting, date-stamping, and drawing inference on the origination and collapse of bubble episodes when prices evolve under persistent, time-varying volatility. Standard bubble tests are typically derived under homoskedasticity or weak heteroskedasticity and may therefore yield misleading inference in more general settings. We extend right-tailed Dickey-Fuller unit root tests to autoregressive models with highly persistent mean and volatility dynamics, delivering a stochastic-volatility-robust ADF (SV-ADF) test that accommodates persistent variance without imposing strict parametric structure. Building on a moderate-deviation asymptotic theory, the SV-ADF yields nuisance-parameter-free procedures with distinct critical values for origination and collapse, producing more stable alarms and fewer transient false positives around volatility spikes. We establish consistency of the date-stamping estimator and show that it remains asymptotically tractable. Monte Carlo simulations document strong power and substantial gains over homoskedastic (PWY) procedures when volatility dynamics are pronounced. An empirical analysis of AI-exposed equities, including the "Magnificent Seven" and leading semiconductor firms, finds pervasive exuberance with substantial heterogeneity in timing, intensity, and duration. The evidence points to especially strong bubble dynamics for Alphabet and TSMC in the current cycle, while Tesla and Nvidia exhibited pronounced explosive episodes in earlier phases of the AI-driven market cycle.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

1 major / 2 minor

Summary. The paper claims to develop a stochastic-volatility-robust ADF (SV-ADF) test by extending right-tailed Dickey-Fuller procedures to AR models with highly persistent mean and volatility dynamics via moderate-deviation asymptotics. It asserts that this yields nuisance-parameter-free critical values with distinct thresholds for origination and collapse, establishes consistency of the date-stamping estimator, demonstrates superior Monte Carlo power over homoskedastic PWY tests under pronounced volatility, and applies the method to AI-exposed equities (including the Magnificent Seven and semiconductor firms) to document heterogeneous exuberance episodes with strong evidence for Alphabet and TSMC in the current cycle.

Significance. If the moderate-deviation theory delivers the claimed nuisance-parameter-free inference and consistency without parametric volatility restrictions, the framework would meaningfully improve bubble detection reliability in markets with time-varying volatility by reducing transient false positives. The Monte Carlo gains and the empirical heterogeneity findings on AI stocks add practical relevance, while the date-stamping consistency result strengthens the methodological contribution for financial time-series analysis.

major comments (1)
  1. [Asymptotic theory section] Asymptotic theory section (description of SV-ADF and moderate-deviation extension): The load-bearing claim that moderate-deviation asymptotics produce nuisance-parameter-free limiting distributions and distinct critical values for origination/collapse, even without strict parametric structure on the volatility process, is asserted but not supported by explicit limiting expressions or proof sketches in the available text; if integrated or near-integrated volatility introduces additional random limits or parameter-dependent normalizations, the robustness and consistency of the date-stamping estimator would not hold as stated.
minor comments (2)
  1. [Abstract and empirical section] Abstract and empirical section: The reference to 'AI-exposed equities, including the Magnificent Seven' would benefit from an explicit list of the firms analyzed and the precise selection criteria used for the sample.
  2. [Monte Carlo section] Monte Carlo section: The volatility processes simulated to represent 'highly persistent' dynamics should be described in more detail (e.g., persistence parameters and whether they remain non-parametric) to allow readers to assess alignment with the theoretical assumptions.

Simulated Author's Rebuttal

1 responses · 0 unresolved

We thank the referee for the careful and constructive review. The single major comment concerns the presentation of the asymptotic theory. We address it directly below and will revise the manuscript to provide greater transparency and rigor in that section.

read point-by-point responses

  1. Referee: The load-bearing claim that moderate-deviation asymptotics produce nuisance-parameter-free limiting distributions and distinct critical values for origination/collapse, even without strict parametric structure on the volatility process, is asserted but not supported by explicit limiting expressions or proof sketches in the available text; if integrated or near-integrated volatility introduces additional random limits or parameter-dependent normalizations, the robustness and consistency of the date-stamping estimator would not hold as stated.

    Authors: We agree that the current exposition of the moderate-deviation asymptotics is too concise. In the revision we will add the explicit limiting distribution of the SV-ADF statistic, derived under the maintained assumptions of persistent stochastic volatility that is itself stationary (though possibly near-integrated). A proof sketch will be supplied showing that the limit is standard normal after suitable normalization and is free of nuisance parameters associated with the volatility process. We will also clarify why the date-stamping consistency result continues to hold without parametric restrictions on volatility and why integrated or near-integrated volatility does not generate additional random limits or parameter-dependent critical values under the moderate-deviation regime. These additions will make the theoretical claims fully explicit and verifiable. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation rests on external moderate-deviation asymptotics

full rationale

The paper claims to extend right-tailed ADF tests via moderate-deviation asymptotics to AR models with persistent mean and volatility, producing nuisance-parameter-free critical values and a consistent date-stamping estimator without imposing parametric volatility structure. No quoted equations or steps reduce the limiting distributions, critical values, or consistency result to fitted parameters or self-defined quantities by construction. The SV-ADF procedure and its distinct origination/collapse thresholds are presented as outputs of the asymptotic theory rather than inputs. Empirical application to AI equities is downstream and does not feed back into the theoretical claims. No self-citation chains, ansatz smuggling, or renaming of known results are exhibited in the provided text that would force the central result. This is a standard case of a self-contained theoretical extension.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

The central claim rests on the applicability of moderate-deviation asymptotics to persistent volatility models and the assumption that the test remains consistent without parametric volatility specification; no free parameters or invented entities are explicitly introduced in the abstract.

axioms (1)
  • domain assumption Moderate-deviation asymptotic theory delivers nuisance-parameter-free critical values and consistency for the date-stamping estimator under highly persistent mean and volatility dynamics.
    Invoked to justify the SV-ADF test and its distinct origination/collapse procedures without strict parametric assumptions on volatility.

pith-pipeline@v0.9.0 · 5578 in / 1409 out tokens · 54734 ms · 2026-05-12T00:56:47.196551+00:00 · methodology

Review history (2 revisions) →

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

5 extracted references · 5 canonical work pages

  1. [1]

    Z., Kindleberger, C

    Aliber, R. Z., Kindleberger, C. P., and McCauley, R. N. (2015).Manias, Panics, and Crashes: A History of Financial Crises. Springer. Astill, S., Harvey, D. I., Leybourne, S. J., Sollis, R., and Robert Taylor, A. (2018). Real-time monitoring for explosive financial bubbles.Journal of Time Series Analysis, 39(6):863–891. Baillie, R. T., Bollerslev, T., and ...

    work page arXiv 2015

  2. [2]

    Engle, R. F. and Bollerslev, T. (1986). Modelling the persistence of conditional variances. Econometric Reviews, 5(1):1–50. Fusari, N., Jarrow, R., and Lamichhane, S. (2025). Testing for asset price bubbles using options data.Journal of Business & Economic Statistics, 43(4):1–15. 24 Gormsen, N. J. and Koijen, R. S. (2020). Coronavirus: Impact on stock pri...

    work page 1986

  3. [3]

    Greenwood, R., Shleifer, A., and You, Y

    Technical report, National Bureau of Economic Research. Greenwood, R., Shleifer, A., and You, Y. (2019). Bubbles for Fama.Journal of Financial Economics, 131(1):20–43. Harvey, D. I., Leybourne, S. J., Sollis, R., and Taylor, A. R. (2016). Tests for explosive financial bubbles in the presence of non-stationary volatility.Journal of Empirical Finance, 38:54...

    work page arXiv 2019

  4. [4]

    n2αδ 2(τ−τ e) n τ 2c2 (τ−τ e) + τX t=τe δ 2(t−τe) n −2 τX t=τe nα τ c δ t+τ−2τ e n # X 2 τe 1 +o p(1) =

    Therefore, by the martingale functional CLT (e.g., Merlev` ede et al., 2019),Mn(·)⇒W(·) inD[0,1], withWstandard Brow- nian motion. Now, we define the right-continuous process Wn t n := 1√n mn,t tX j=1 σjεj, W n(u) :=W n ⌊nu⌋ n , u∈[0,1]. Then, Xt =X 0 + tX j=1 σjεj =X 0 + √n mn,t Wn t n , and,W n(·)⇒W(·) inD[0,1]. Let ¯Xτ := 1 τ τX t=1 Xt−1 =X 0 + √n mn,n...

    work page 2019

  5. [5]

    38 Recursive SV−ADF Statistic Origination − Jan 2021 Collapse − Jun−2021 −5.0 0.0 5.0 10.0 15.0 20.0 25.0 30.0 35.0 0 0.5 1 1.5 2 2.5 3 2020 May 2020 Nov 2021 May 2021 Nov 2022 May 2022 Nov Cardano −40.0 −20.0 0.0 20.0 40.0 60.0 80.0 100.0 120.0 0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 2020 May 2020 Nov2021 May 2021 Nov2022 May 2022 Nov Dogecoin Origination − Aug 20...

    work page 2021