Santiment Research Papers
Academic-style research papers from Santiment Research, evaluating Santiment data as inputs to trading strategies, risk-state features, and attention baselines.
Full reports are available to institutional subscribers; the first page of each paper is available below.
Social Data as a Strategy Enhancement Layer
Cost-Aware Evidence from Three Matched Crypto Strategies
Author: Larry Cao, Santiment Research
This paper tests whether lagged Social Volume and Social Dominance improve three daily BTC, ETH, and SOL strategies, each compared against the same market rule with the Social input removed. Signals used one-day-lagged Social observations, executed on the next day’s return, with 5 bps of cost per position or weight change. Parameters and portfolio weights were fixed from 2022–2024 development data before the retrospective test.
Across the full sample, the Social-informed portfolio achieved a 14.5% CAGR, a 0.79 Sharpe ratio, and a -26.1% maximum drawdown, against 4.2% CAGR, a 0.31 Sharpe ratio, and a -34.6% maximum drawdown for the matched market portfolio. The advantage held in every reported calendar partition except the final 2026 partition, which was approximately flat.
Full report access available for institutional subscribers.
First page preview Santiment Social Volume Intraday Seasonality
A Baseline for Crypto Attention
Author: Larry Cao, Santiment Research
This paper measures hour-of-day and weekday seasonality in Santiment Social Volume for ten large-cap assets — bitcoin, ethereum, solana, xrp, dogecoin, binance-coin, cardano, chainlink, avalanche, and tron — over complete UTC days from 2025-06-20 through 2026-05-19.
The paper’s conclusion is a practical one: treat hour-of-day and weekday seasonality as the first risk control in any short-horizon Social Volume strategy, since raw data is not a clean abnormality signal unless the clock is part of the model. This is framed as an attention-baseline study, not a claim about return predictability.
Full report access available for institutional subscribers.
First page preview Santiment Data Anomalies as Conditional Market-State Features
A Production-Audited Event Study on Binance-Tradable Crypto Assets
Author: Santiment Research
This paper evaluates four Santiment data anomalies — Social Dev Score, ETH Whale Dump, Price/Network Activity Divergence, and Project in Trends — as conditional market-state features. Production event rows are compared against same-asset, same-quarter control timestamps over 1h, 4h, 24h, and 72h horizons, using Binance hourly USDT prices for 1,211 unique events across 34 mapped tradable assets from 2023-01-01 through 2026-05-28.
The strongest evidence is concentrated in Price/Network Activity Divergence, which shows positive event-minus-control signed returns and materially higher realized volatility. Project in Trends behaves more like an attention and risk-state marker: post-event realized volatility rises while 24-hour signed returns are negative in this sample. ETH Whale Dump remains an ETH-specific event-risk label, and Social Dev Score has limited statistical evidence after the Binance and sparse-asset filters.
The paper frames these as conditional event-state evidence, not a transaction-cost-aware trading strategy, out-of-sample validation, or standalone proof of alpha.
Full report access available for institutional subscribers.
First page preview Social Heating and Forward Returns
Evidence from Historical Binance Market-Cap Universes
Author: Larry Cao, Santiment Research
This paper tests whether unusually high and rising social attention (“social heating”) signals stronger subsequent returns, using monthly top-100 Binance USDT spot universes ranked by historical market capitalization, excluding equities, stablecoins, and leveraged tokens, with signals spanning January 2023 through September 2026. At the baseline heating threshold, 1,249 evaluable events had a mean seven-day return of 1.81%, against 0.84% for same-date, momentum-matched non-heating peers.
Mean excess was positive at all seven tested z-score cutoffs (z ≥ 2 through z ≥ 10), with mean returns ranging from 1.81% to 3.90% and excess returns ranging from 0.42 to 1.31 percentage points. Median excess was negative at every cutoff, reflecting a distribution shaped by occasional large winners rather than a consistent edge. Greater extremity did not reliably strengthen the advantage: the clearest evidence sits at moderate heating levels, not the most extreme ones.
The paper frames this as a positive historical association between social heating and average forward returns relative to the specified peers, not evidence of real-time profitability — historical data revisions, incomplete coverage, and unmodeled trading costs limit that interpretation, and net trading profitability still needs to be tested.
Full report access available for institutional subscribers.
First page preview