The methodological spine of the portfolio

The Research Engine

A unified long-memory + multi-scale + information-theoretic + network-formation substrate. Each framework asks one scientific question by parameterising the substrate on a specific panel, calendar, and channel taxonomy.

SHSSM · IIT Bhubaneswar.

contacting the live engine…

One substrate · eight shared primitives

Every framework below is built from the same toolkit. Nothing in the portfolio is bespoke at the method level — what changes between papers is which primitives are composed, on which panel, at which calendar and quantile.
Long-memory estimation DFA · GPH · Hurst
Measures how persistent a return series is — the Hurst exponent that drives market-efficiency node-weights.
Multi-scale wavelet decomposition MODWT · LA8
Splits each series into time-scale bands (days → weeks → months) using the maximal-overlap discrete wavelet transform, Daubechies LA8.
Transfer entropy binned · KSG · quantile
Directional information flow X→Y — estimated three ways: histogram-binned, KSG k-nearest-neighbour, and quantile-conditioned for tails.
Surrogate inference IAAFT · block-bootstrap
Builds null distributions that preserve a series' own structure, so a measured spillover is tested against chance, not asserted.
Community detection Leiden · Walktrap · Louvain · CNM
Finds clusters of markets that move together inside the spillover network — the mesoscale structure of contagion.
Network-formation game theory CARA-Nash · CRRA-Bellman
Treats the network itself as an equilibrium outcome of agents trading off risk and connection, not just a correlation snapshot.
Structural channel attribution IV/2SLS · LASSO · local projections · Rigobon
Assigns each contagion link to an economic mechanism (trade, financial, policy…) using identification-by-heteroskedasticity and instruments.
ROC validation DeLong · Youden
Scores how well a risk index separates crisis from calm — area-under-curve with DeLong tests and the Youden optimal threshold.

Six frameworks · one substrate

Each card is a single scientific question, the substrate parameterisation that answers it, the current publication status, and one headline result. Two frameworks are on arXiv; four are working papers or in preparation with no DOI — and we say so.
NAMH working paper · 2026
Does network-mediated adaptive efficiency explain Hurst-exponent crisis dynamics?
DFA-Hurst + KSG transfer entropy + IAAFT surrogates + HWTE weighting + Leiden communities + network fixed point.
Headline resultJapan the eigenvector hub in 13 of 20 windows (magnitude-gated; the surrogate-FDR network is empty — see Reproduce). Hurst panel reproduces live to floating-point precision (max Δ < 1e-8).
see it reproduce live →
Frontiers III working paper · 2026
Is the market for news estimable — can we measure the directed network of news attention that the theory says veracity depends on?
Log-change stationarisation + directed KSG / Frenzel–Pompe transfer entropy (max-norm) + IAAFT source surrogates + Leiden communities + PageRank / Katz centralities.
Headline resultInflation broadcasts, election is the dominant attention sink; a topic's connectedness is positively associated with the attention it commands (ρ = +0.67, p = 0.007). The directed transfer entropy reproduces live to the published precision — TE(inflation→election) = 0.065781 — while the surrogate-FDR network is empty, reported honestly (see Reproduce).
see it reproduce live →
MCPFM arXiv:2507.08065
Can a protocol-based agent-communication game yield a tractable systemic-risk index?
MODWT + binned transfer entropy + block-bootstrap + Walktrap communities + SRI composite + ROC/DeLong validation.
Headline resultSystemic-Risk Index AUC 0.915 (US, COVID-19 crash) — published; live regeneration pending.
arXiv:2507.08065 ↗
contagion-channels arXiv:2604.26546 · CRAN v0.1.3
What share of cross-border equity contagion is Trade / Financial / Geopolitical / Behavioural / Monetary-Policy?
MODWT + quantile transfer entropy + structural IV/2SLS attribution + Cinelli-Hazlett sensitivity.
Headline resultFinancial dominates the European Sovereign-Debt Crisis (39.5%, identified); the GFC Trade share (27.9%) is flagged exploratory (Sargan).
arXiv:2604.26546 ↗
WaveQTE working paper
An earlier activation-scoring approach to the same channel question — what changes when hand-crafted scores are replaced by structural-IV identification?
GPH long-memory + MODWT + quantile-regression transfer entropy + Walktrap; Stage-2 activation scoring (18 markets, 2004–2024, 7 crisis windows).
Headline resultActivation scoring puts Trade at 3.9% of GFC links; the structural-IV upgrade (contagion-channels) reverses this to 27.9% — identification changes the conclusion.
working paper · no DOI · direct predecessor of contagion-channels
commodity in preparation
The same question for 23 commodity futures rather than equities.
DFA-Hurst + quantile transfer entropy + Clauset-Newman-Moore communities.
Headline result— results not yet released
coming soon
Plus SOCH — the Scale-Ordered Contagion result (arXiv:2606.04113, the published sochcontagion package) is the portfolio's flagship live reproduction: it regenerates on real G20 data, scale by scale, on the Reproduce page — max Δ < 0.004, and the scale-ordering prediction holds (p = 0.042).

What an evaluator can check today

Most of these are fixed facts of the published record. One is read live from the compute engine on page load — the number of registered methods.
5
frameworks
3
arXiv preprints
3 + 1
R packages (public + pending)
methods live
How to read this: the three arXiv preprints are MCPFM (2507.08065), contagion-channels (2604.26546) and the Scale-Ordered Contagion result (2606.04113). The R packages are the group's published code — contagionchannels (CRAN v0.1.3) · ManyIVsNets (CRAN v0.1.1) · sochcontagion (r-universe v0.1.0) — with a 4th, namh, release-pending. The "methods live" tile is fetched from the engine catalog on load; if the engine is unreachable it falls back to the last known count, labelled as such.

Cross-paper consistency — divergent by design, shared at the spine

A common substrate does not mean identical settings. Where the frameworks differ, they differ on purpose, because they ask different scientific questions. Where they agree, the agreement is the methodological spine that makes them one engine.
Methodological marker Per-framework setting Status
Rolling-window length NAMH 252d · MCPFM 21d · commodity 260d Divergent — different questions
Transfer-entropy estimator NAMH KSG k-NN · MCPFM binned · contagion-channels quantile-conditioned Divergent by design
Surrogate count B NAMH 200 · MCPFM 200 · contagion-channels 1000 Divergent
Community detection NAMH Leiden · MCPFM / contagion Walktrap · commodity CNM Divergent
Multi-scale basis All MODWT Daubechies LA8 Shared / consistent
Stationarity discipline All transfer entropy on log-returns I(0), never price levels I(1) Shared / consistent
Divergence here is a deliberate scientific choice, not an inconsistency: a 21-day systemic-risk window and a 252-day efficiency window answer different questions on purpose. The shared rows — the same wavelet basis, the same I(0) stationarity discipline across every paper — are the common spine that makes these five frameworks one engine. Status is conveyed by the text label (and an amber/green marker), never colour alone.

Compute economics & reproducibility

What it costs to run, and how anyone can verify the results without taking our word for it.

Compute economics

₹4,50,000 in Google Cloud research credits

Funded by Google Cloud research credits awarded to SHSSM, IIT Bhubaneswar, expiring April 2027. This covers the sandboxed compute engine that powers the live reproductions and the Workbench.

Live billing meter: not yet wired. We do not display a precise spend or compute-hours figure because it is not yet metered — when it is, it will appear here as a measured number, not an estimate.

Reproducibility ledger

Three published surfaces regenerate live on real data through each paper's own published package — so those live and published numbers are identical by construction. Two heavier surfaces (the MCPFM systemic-risk index and NAMH's surrogate-gated network) are honestly marked pending, not simulated.

Open the reproducibility ledger →

Every result in the Workbench carries a permalink, a CSV/JSON export, and a one-click BibTeX citation — a number becomes a citable artifact.

All code on GitHub ↗

One substrate · five frameworks · one live engine.
The live method count is read from the sandboxed compute engine (shssm-compute…run.app) on page load. On arXiv: MCPFM (2507.08065) · contagion-channels (2604.26546) · SOCH (2606.04113). NAMH, WaveQTE and commodity are working papers / in preparation with no DOI.
Reproduce the papers · Workbench · GitHub · SHSSM, IIT Bhubaneswar.