Can Our Model Tell ARK's Winners from Its Blowups?

An honest backtest — including where our model failed.

4 min read · 20 companies · Two blind tests · Not investment advice

The question

In February 2021, ARK Invest stood at the peak of its influence, making a bold, concentrated bet on disruptive innovation. We support that challenge. And it gave us a perfect laboratory: could the Blitz Index — our 100-point model of blitzscaling strategy — have told ARK's future 10-baggers apart from its blowups?

We took ARKK's top 20 holdings as of February 26, 2021, scored each one blind using only information available at the time, and measured what happened over the next 5.7 years. Then we ran the whole thing again at the December 2022 trough. Same 20 companies, different valuation regime.

Test 1: The peak — our hypothesis was rejected

Blind date: February 28, 2021. Forward window: February 2021 → October 2026. The correlation between blind scores and forward returns was r = −0.15 — effectively zero. The highest-scoring third of companies averaged −52.0%. The model could not pick winners at the peak.

But one thing worked. Of the 7 deep blowups (down 70% or more), the model avoided 5 with sub-65 scores — including Invitae (49/C → bankrupt, −100%) and Teladoc (51/C → −97.4%).

Test 2: The trough — the mirror image

Blind date: December 31, 2022. Same 20 companies. The correlation flipped to r = +0.31 — weak, but in the predicted direction. And here is the striking part: the average blind score fell 8.4 points (businesses had genuinely deteriorated), while average returns went from −27.8% to +151.5%.

Honesty requires the other side too: the biggest winners at the trough — Spotify (+570%), Twilio (+491%), Twist Bioscience (+571%) — were all mid-tier C grades. The model did not predict the magnitude ranking.

What we learned

On ARK's challenge

This research was conducted in support of ARK's challenge, not against it. Concentrated bets on disruptive innovation are a hard road, and the Blitz Index read the business substance of ARK's portfolio substantially correctly — good businesses as good businesses, doomed ones as doomed. In February 2021, the market sorted winners and losers by price, which sits outside this model's measurement range. That is the next question for the theory, and ARK's challenge remains open.

Limitations

• n = 20 — statistical power is limited; r = +0.31 is directional evidence, not proof.
• Some starting prices are estimates (disclosed as ranges in the full report); they do not change any positive/negative classification.
• One universe, two dates. February 2021 was a historic valuation peak — that specificity likely dominates the result.
• Past validation does not predict future results.