How the research works.

The edge

Markets are usually net-long, so perp longs pay shorts. Short the perp, hold spot as the hedge, collect the funding. Structural, not a pricing error — and it turns negative when the market flips.

Charged, not assumed away

  • Fees

    Both legs, every trade.

  • Slippage

    No fills at mid.

  • Funding, both ways

    Paid when rates turn.

  • Rebalancing drag

    Neutrality has a turnover cost.

Rebalance daily and costs take the edge. Weekly or biweekly keeps it.

Bias checks

  • Look-ahead

    Decision-time data only.

  • Survivorship

    Liquid majors, not hindsight winners.

  • Overfitting

    Coarse parameters; random data shows ~zero edge.

  • Out-of-sample

    Tested on an unseen segment.

The forward record

  • Daily at 00:30 UTC, one point per day, same five majors and parameters as the backtest.
  • No data, no point. Nothing is interpolated or filled; the run answers HTTP 503. (A silent 200 once froze the record for 22 days.)
  • Binance first, Bybit fallback — never blended; the source is named per day on /paper.
  • Stale after 48 h: /paper and /funding say so, computed at render time, failing safe.
  • No live track record yet. Zero real fills; /track-record stays empty until one exists.

Check the files

public/results.json/results
public/paper.json/paper
public/live-trades.json/track-record
public/funding-fallback.json/funding
See the backtest →