TL;DR: The long/short ratio and on-chain or social sentiment describe how the crowd is positioned and how it feels, not where price is going. They work best as filters that keep a price-based strategy out of crowded or euphoric conditions. Calibrate thresholds to how each series actually behaves (the aggregated accounts long/short ratio is almost never below 1), align daily sentiment point-in-time, and remember that the history is short.
What do the long/short ratio and sentiment data measure?
These series all try to capture the state of other traders, as opposed to price itself.
- Long/short ratio: the number of accounts net long divided by the number net short on a market, aggregated across exchanges. It is a headcount, not a dollar amount.
- Social volume: how much a coin is being discussed across social and community channels.
- Social sentiment: whether that discussion leans positive or negative.
- On-chain activity: exchange-balance changes (coins moving onto or off exchanges), the number of large transactions, active addresses, development activity, and valuation ratios such as MVRV (market value to realized value).
The appeal is intuitive. Crowds tend to be most confident near local tops and most fearful near local bottoms. Positioning data can show when one side is overcrowded and vulnerable to a squeeze. The catch is that crowds can stay crowded for a long time while price keeps moving their way.
Why use sentiment as a filter rather than a signal?
Sentiment and positioning data are slow, noisy and often coincident with price rather than leading it. Social volume spikes when price moves sharply; that is mostly a description of the move, not a prediction of the next one.
Used as a filter, the same data does a more modest and more defensible job: it vetoes trades in conditions where your core idea tends to fail.
| Data | Contrarian filter idea | Confirmation filter idea | Main caveat |
|---|---|---|---|
| Long/short ratio | Skip new longs when the crowd is unusually long | Allow longs when the ratio is low for that asset | Long-biased; levels differ by asset |
| Social volume | Avoid entries during attention spikes | Require rising attention for breakout trades | Confirms moves; rarely leads |
| Social sentiment | Avoid longs when sentiment is strongly positive | Allow longs only when sentiment is not negative | Unbounded provider metric; noisy |
| Exchange-balance delta | Treat inflows as potential sell pressure | Treat outflows as accumulation | Native-coin units; exchange labeling is imperfect |
| Whale transaction count | Caution when large transfers surge | Confirm moves with rising large-transfer activity | Activity, not direction |
| MVRV | Avoid longs when holders sit on large unrealized gains | Favor longs when MVRV is low | Very slow; a regime tool, not timing |
Pair each of these with a price-based trigger. The no-code strategy builder guide covers the trigger, filter, signal and risk structure; the sentiment series belongs in the filter slot.
Why is the crypto long/short ratio almost always above 1?
This is the most common mistake with the long/short ratio, and it produces strategies that never trade.
The series Trigr uses is the aggregated accounts ratio. Because it counts accounts rather than dollars, and retail accounts in crypto lean long, it is structurally long-biased. Across BTC, ETH, SOL and DOGE it has typically ranged between roughly 1.5 and 4 and has almost never gone below 1.
So a filter like "only go long when the ratio is below 1, meaning the crowd is net short" sounds sensible and never fires. The backtest shows zero trades, or a handful from a data glitch. Trigr's default for this node is a ratio below 2, which marks unusually light crowd longs at a level that actually occurs.
Practical rules for thresholds:
- Calibrate per asset. A "high" ratio on BTC may be ordinary on a smaller coin. Look at the series before choosing a level.
- Prefer relative conditions. "Ratio below 2" or "ratio above 3" relative to its usual band is more meaningful than a theoretical midpoint of 1.
- Pair it with funding. Funding shows whether longs or shorts are paying to hold positions, which is a dollar-weighted view of crowding. Hyperliquid's funding documentation explains the mechanics, and the funding-rate and open-interest guide shows how to combine the two.
How do you align daily sentiment data without look-ahead?
Most on-chain and social data is published daily. If your strategy trades on 1H bars, you need to decide which daily value is known at each hour. The honest answer is the last completed day. Using "today's" sentiment at 10:00 means reading a value that summarizes the whole day, including the 14 hours that have not happened.
Trigr's engine applies the same point-in-time rule to every node: a node reads only data available at the bar's close, and a coarser series is carried forward from its last completed value, never read ahead. Orders then fill at the next bar's open. The look-ahead bias guide shows how easily a daily join leaks a day of information into an intraday backtest.
A second, subtler issue is revision. Some on-chain metrics are recalculated as data providers relabel exchange wallets or backfill. A series that looks clean in history may have looked different on the day. Treat any strategy that depends on a precise on-chain threshold with extra suspicion.
Which sentiment and positioning sources does Trigr support?
In Trigr's Studio these are API data nodes, alongside indicators and other derivatives data. At the time of writing:
- Long/short ratio: hourly, aggregated accounts ratio, with roughly three months of real history. Usable as a trigger or filter on supported crypto perps.
- Santiment social: social volume (rising or falling over a lookback, filter only) and signed social sentiment. Daily, with about a year of history.
- Santiment on-chain: exchange-balance delta (net inflow or outflow, in native coin units), whale transaction count (transactions of at least $100k, rising or falling, filter only), MVRV, active addresses and development activity. Daily, for a set of supported assets.
Santiment's academy documents how each metric is defined. Coverage differs by asset; the live data catalog and the capability catalog show which series exist for which market. TradFi markets on Hyperliquid HIP-3 have OHLCV-only inputs in the current beta, so none of these apply to GOLD, SPX or stock perps.
What does an example look like?
An idea to test, not a recommendation. It trades SOL on 4H bars.
- Trigger: SOL 4H SuperTrend flips up.
- Filter: SOL long/short ratio < 2.5 (the crowd is not heavily long).
- Filter: SOL social volume not rising over 7 days, expressed as "falling" (avoid attention spikes).
- Signal: long when all three hold.
- Risk: 8% size, 2x leverage, long only, ATR trailing stop, 5% stop-loss, exit on signal flip.
The hypothesis is simple enough to state in advance: trend entries work better when the crowd is not already positioned for them. Whether the data supports it is what the backtest is for.
How should you evaluate a sentiment-filtered strategy?
The short history of these series is the dominant fact. Three months of hourly long/short data, or a year of daily Santiment data, will produce few trades for most strategies. With a small sample, a filter can look brilliant by removing two or three bad trades.
- Compare with and without each filter, net of costs. Fees and the builder fee are always applied; slippage and funding are opt-in and should be on.
- Count trades removed. If the filter removes most trades, the remaining sample is too small to judge.
- Watch for zero-trade runs. They usually mean a threshold outside the series' real range, as with a long/short ratio below 1.
- Keep the trial count visible. Trying many thresholds and lookbacks is the fastest route to a lucky result. Iterate as labeled experiments inside one strategy, and read the selection bias guide before trusting the best variant.
- Forward test. Freeze the strategy and run it on a paper agent to build a record that no backtest can provide.
A plain backtest does not compute the Deflated Sharpe Ratio or the Probability of Backtest Overfitting; an ML optimization run reports both, plus feature importance that can show whether a sentiment feature carries any information. The shaded trailing 25% in Studio is a recent-period diagnostic, not an out-of-sample holdout.
What this means for you
You can add crowd positioning and on-chain context to a strategy in a node or two, with point-in-time alignment handled for you and defaults calibrated to levels the data actually reaches. The same graph runs in the backtest, on a paper agent and on a live Hyperliquid agent. The sentiment series used here are certified for live evaluation as well as backtests, so a filter does not silently change behavior when you deploy.
Backtests are not guarantees; perps are leveraged and can lose more than expected.
Next steps
Look up the series available for your market in the capability catalog, then add one positioning filter to a strategy you already trust and compare net results. You can also have an assistant do it: connect ChatGPT, Claude or Codex over MCP and ask it to test the filter as an experiment.