Are cryptocurrencies the future of fraud?

Bennett Kleinberg

Dawes Centre for Future Crime, UCL

Anatomy of a P&D

The three phases of a pump-and-dump operation (Kamps & Kleinberg, 2018)

Two important events

  1. The Internet happened
  2. Cryptocurrencies

What could go wrong?

Crypto P&Ds

Old challenges New challenges
low market cap faster time scale (near real-time)
lack of reliable information broader spread of misinformation
low price new vehicles for “rumours”

P&D in the wild

Can we detect them automatically?

Data

  • scraped cryptocurrency exchange data
  • trading data of 1-h granularity
  • 20 days of trading, 5 exchanges, 977 trading pairs

Data + code available at https://osf.io/827wd/

Criteria for P&Ds

  • sudden price increase (PUMP)
  • sudden volume increase (PUMP)
  • marked price drop (DUMP)

Conditional local point anomaly detection

Anomaly parameters

Balanced parameters
Estimation window 12h
Volume increase 300%
Price increase 5%
Price drop 1.00 SD
Alleged pumps 2150
Pump-and-dumps 1617
Crypto/crypto 97.0%
Low market cap 81.8%

P&D detection

Core findings

  • potential for detectability
  • mainly crypto/crypto pairs
  • mostly low market cap

Zooming in

  • Exchange-level
    • % of traded coins \(\neq\) % of P&Ds
    • Some exchanges are used more often than others
    • e.g. “Kraken”: 6% of traded coins, < 1% of P&Ds
    • among most regulated marketplaces

Zooming in

  • Coin-level
    • Most are never targeted
    • Some are targeted again and again
    • 30% of coins \(\sim\) 80% of P&Ds

Repeat victimisation of coins

Core findings

  • potential for detectability
  • mainly crypto/crypto pairs
  • mostly low market cap
  • repeat exchange victimisation
  • repeat coin victimisation

Outlook & bigger picture

New questions

  • What are the characteristics of repeat victims?
  • How is misinformation spread?
  • How are groups building trust?
  • What’s more than pump-and-dumps?

Ongoing work

  • Detection:
    • Signals in the accumulation phase
    • Signals in the pumping phase
  • Anticipation:
    • Changes to MO
    • Transition to new targets

Interdisciplinary challenge!

Bigger picture

If you had to design policing from scratch …

  • fully embrace data
  • avoid catch phrases (“big data”, “AI”)
  • solve the capabilities problem
  • go “open data”

Worst case: the gap becomes so wide, we’re not realising there is one.

Are cryptocurrencies the future of fraud?

Yes Maybe No

Thank you

Collaborator: Josh Kamps

bennett.kleinberg@ucl.ac.uk // bkleinberg.net