Why on-chain growth didn’t make the cut.
Real network usage growth is the most appealing “fundamental” a crypto factor could have — and genuinely independent of funding. It still didn’t reach our bar. Here’s the honest result, including two bugs that briefly fooled us.
A real fundamental, and price-neutral
If a chain is genuinely being used more — more active addresses, more transactions — that adoption should eventually matter. Crucially, we measured the growth of active addresses, not the level, which keeps it independent of price (rising usage isn’t just “number went up”). And its cross-sectional correlation with funding crowding was near zero. A genuinely different axis — exactly what a real second factor needs to be.
Two self-inflicted false alarms
Honesty means owning your own mistakes, and this factor produced two:
- An early, exciting t-stat came from a data fetch that silently truncated on rate-limit errors — it was computed on partial data. Fixed to fail loudly instead of truncating, the signal shrank.
- A later “significant” result was overlap-inflated: a 14-day signal measured on overlapping windows, which violates our own non-overlap rule. On clean windows it faded.
Both were caught because independent checks disagreed with each other — impossible from correct code, so a bug had to exist. The keeper lesson: never trust a t-stat until you’ve verified data completeness and non-overlapping windows.
A faint tilt, below the bar
On the clean, non-overlapping test the signal came in at t ≈ 1.80 (p ≈ 0.07) — pointing the right way (rising usage → modestly higher forward return) but below our significance bar. The tradeable portfolio was flat and regime-fragile (it only did anything in downturns), and the whole result was scoped to ~12–14 older coins, missing newer high-activity chains. Not enough to ship.
Parked, not buried
On-chain growth is parked as unproven — the honest verdict is “insufficient evidence,” not “dead” and not “a manufactured signal.” It’s the candidate most likely to earn a second look.
What would make us reconsider
- More statistical power — more coins (including the newer high-activity chains) and longer history.
- A clean, non-overlapping result that clears significance and adds over funding crowding.
Common questions
Is on-chain address growth correlated with funding?
Barely — its cross-sectional correlation with funding crowding was near zero, and it survived as an independent axis even after controlling for funding and momentum. That’s what made it worth chasing. It failed on strength, not on redundancy.
Why do you call it "unproven" rather than "dead"?
On a clean, non-overlapping test the signal came in at t ≈ 1.80 (p ≈ 0.07) — a faint tilt in the right direction, but below our significance bar. The portfolio was flat and regime-fragile. That’s genuine insufficient evidence, not a manufactured signal and not a definitive failure.
What were the bugs you caught?
Two. First, an early strong t-stat came from a data fetch that silently truncated on rate-limit errors — partial data inflated the result. Second, an apparent significance was overlap-inflated (a 14-day signal measured on overlapping windows). Both faked an edge; both were caught by cross-checks. The lesson: always verify data completeness and use non-overlapping windows before trusting a t-stat.
Ioxer is research, not investment advice. IOX is a crowding read — not a price prediction, not a buy/sell signal.