Quant in Crypto Land
Jesus Rodriguez
IntoTheBlock
QUANT IN CRYPTO LAND
ITB Analytics Platform
● Over 150 customers that are currently distributing its data under different integration models.
● Analyzing over 50TB of data
● Leveraging machine learning to derive actionable insights about crypto assets
FINTECH AND
NEWS SITES (45)
WALLETS (5)
DATA
PROVIDERS (9)
BLOCKCHAIN
PROJECTS (5)
EXCHANGES (35)
Quant Intelligent Services Unit Traction
● ITB has partnered with the top quant funds in the industry to provide alpha-
generating intelligence.
● Over $500M deployed in ITB crypto quant strategies
● ITB receives a % of the profits generated by each quant fund’s strategy that
uses ITB services.
QUANT IN CRYPTO LAND
Agenda
● Unique characteristics of quant strategies in crypto
● Unique sources of alpha
● DeFi as a new frontier for HFTs
● Inefficiencies in crypto
● NFTs as a new form of asset class
● Why most crypto quant strategies fail?
● Cutting edge deep learning and crypto quant strategies
QUANT IN CRYPTO LAND
Some factors that makes crypto unique for
quant strategies...
QUANT IN CRYPTO LAND
Unique Sources
of Alpha
New Programmable
Financial Primitives
Regularly Inefficient
Market
Four Core Factors
New Types of
Digital Assets
1 2 3 4
Unique Sources of Alpha…
QUANT IN CRYPTO LAND
Blockchain datasets represent a very unique
dataset to understand the behavior of groups
of investors…
QUANT IN CRYPTO LAND
Some ideas?
QUANT IN CRYPTO LAND
Support and Resistance Based on Wallet Holdings
Ex: Bitcoin
QUANT IN CRYPTO LAND
Event Signals Based on Flows Into Exchanges
Ex: Bitcoin
QUANT IN CRYPTO LAND
Or Miners
Ex: Bitcoin
QUANT IN CRYPTO LAND
Patterns About Holding Time, Geography of
Relationships of Individual Investor Positions
Ex: Bitcoin
QUANT IN CRYPTO LAND
And many more...
QUANT IN CRYPTO LAND
Blockchain datasets can provide a unique set
of features to quant models that can lead to
robust alpha capture…
QUANT IN CRYPTO LAND
New Financial Primitives…
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Decentralized finance(DeFi) enables the representations of financial primitives as
decentralized programmable smart contract…
Lending Market Making Capital Formation
Asset Composition Insurance Derivatives
Programmable, on-chain, financial primitives
represent a new frontier for high frequency
trading…
QUANT IN CRYPTO LAND
The DeFi HFT Perspective
QUANT IN CRYPTO LAND
Block Time Speed Factor
Trade Transparency Factor
Cost Factor
MEV Factor
The DeFi HFT Perspective
QUANT IN CRYPTO LAND
Block Time Speed Factor
Trade Transparency Factor
Cost Factor
MEV Factor
The HFT DeFi
Transactions are limited by the processing
time of each block
The Result
Speed advantage is relatively limited
The DeFi HFT Perspective
QUANT IN CRYPTO LAND
Block Time Speed Factor
Trade Transparency Factor
Cost Factor
MEV Factor
The HFT DeFi
All trades are visible to all competing
parties
The Result
Every single HFT trade can be copied or
subject to competition
The DeFi HFT Perspective
QUANT IN CRYPTO LAND
Block Time Speed Factor
Trade Transparency Factor
Cost Factor
MEV Factor
The HFT DeFi
There is a cost associated with
transaction execution
The Result
Cost becomes an economic factor
evaluating the viability of specific trades
The DeFi HFT Perspective
QUANT IN CRYPTO LAND
Block Time Speed Factor
Trade Transparency Factor
Cost Factor
MEV Factor
The HFT DeFi
Trades are dependent on miner’s criteria
for placing transactions in a block
The Result
Perfectly viable HFT trades can fail
because of the mining criteria
Regularly Inefficient Market…
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Information Asymmetry
Insiders in crypto projects have
regular access to material,
market influencing information
News Overreaction
News in crypto cause
dispositional movements in the
market
Massive Event Driven
Trades
New token listings, project
updates…
Crypto Market Inefficiencies
CeFi-DeFi Arbitrages
Asset prices in DeFi and CeFi
exchanges are regularly
influenced by different factors
Cross Protocol Arbitrages
Lending, borrowing, market making
dynamics are drastically different
across different protocols
Some examples...
QUANT IN CRYPTO LAND
Crazy Correlations with Twitter Sentiment
Ex: Cardano
QUANT IN CRYPTO LAND
DeFi Borrow-Lending Rates Arbitrage
QUANT IN CRYPTO LAND
New types of digital assets…
QUANT IN CRYPTO LAND
Non-Fungible tokens are a typical case
of a crypto-based asset class that
trades based on different dynamics than
the underlying platform
QUANT IN CRYPTO LAND
NFTs
QUANT IN CRYPTO LAND
Inefficient Indexes
QUANT IN CRYPTO LAND
Crazy Volume
QUANT IN CRYPTO LAND
Challenges of Crypto Quant
Strategies…
QUANT IN CRYPTO LAND
Most crypto quant strategies run
into very peculiar challenges…
QUANT IN CRYPTO LAND
Challenge #1
Small datasets
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Datasets are really small
compared to traditional
capital markets
Crypto is Different
Most ML quant models
struggle learning with small
datasets
The availability of existing
high quality datasets is very
unstable
Challenge #2
Regular “outlier” events
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Outlier events happen
regularly(ex: Bitcoin recent
rally, March crash, the raise
of DeFi…)
Crypto is Different
ML quant models struggle to
learn patterns they haven’t
seen in training datasets
Some outlier events
introduce noise in training
datasets
Challenge #3
The regular retraining dilemma
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Outlier events cause ML
crypto quant strategies to fail
and force them to
incorporate new records in
the training dataset
Crypto is Different
Training ML crypto quant
models with the latest data is
a recipe for overfitting
Outlier training records could
distort the behavior of crypto
quant strategies
Challenge #4
Data quality and reliability
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Crypto datasets are plagued
with “bad data”(ex: wash
trades, fake volume…)
Crypto is Different
Crypto data providers are
unstable and fail all the time
ML crypto quant models fail
with bad or missing data
Challenge #5
Anonymous blockchain records
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Anonymous blockchain
records introduce a
misleading level of uniformity
in training datasets in ML
crypto quant models
Crypto is Different
There is no consistency in
the way crypto exchanges
interact with the underlying
blockchains
There is no linear correlation
between blockchain activity
and trading activity
Challenge #6
Simple model fallacy
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Crypto is too inefficient for
simple ML quant strategies
Crypto is Different
It’s hard to model the
behavior of crypto assets
with a handful of parameters
Quant research based on
simple ML models results
impractical in the crypto
space
Some cutting-edge ideas that might
influence quant strategies in crypto…
QUANT IN CRYPTO LAND
Crypto is coinciding with the
renaissance of deep learning methods
which can have a disproportional impact
in quant strategies in the space…
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
Graph Neural Networks
Blockchain data is fundamentally
hierarchical and well suited for
GNN models
Transformers for Time Series
Traditional time series forecasting models will
struggle with the complexities of the crypto market.
Transformer models applied to time series datasets
can have strong predictive capabilities
Self-Supervised Learning
The anonymity of blockchain datasets it’s a challenge
for most supervised learning methods.
Self-supervised learning techniques can organically
learn patterns using unlabeled blockchain datasets
Crypto Market Inefficiencies
Reinforcement Learning for On-
Chain Data
The on-chain transparency of DeFi trades makes it
a good candidate for reinforcement learning
models
Some practical lessons about building
quant strategies in crypto…
QUANT IN CRYPTO LAND
QUANT IN CRYPTO LAND
The Simplest Trades Work
Market inefficiencies cause very simple trading
strategies to work in crypto Ex: basis trading.
Assume Very Fragile Quant
Infrastructures
Poor quality datasets, regular exchange failures.
Traditional Quant Strategies can be
Adapted to Crypto
Many traditional order book or derivative strategies
can be retrained for crypto assets.
Crypto Market Inefficiencies
Plenty of Opportunities for New
Strategies
DeFi, new derivatives open the door to new types
of strategies.
Summary
● Crypto represents a new landscape for quant strategies
● The immaturity of the crypto space creates very unique challenges for quant strategies
● DeFi represents a new frontier for HFT quant strategies
● NFTs are a great example of new trading dynamics
● New deep learning methods could have an impact in crypto quant strategies
QUANT IN CRYPTO LAND
Thank You!
Jesus Rodriguez
jr@intotheblock.io
medium.com/@jrodthoughts
intotheblock.com

Quant in Crypto Land

  • 1.
    Quant in CryptoLand Jesus Rodriguez IntoTheBlock
  • 2.
    QUANT IN CRYPTOLAND ITB Analytics Platform ● Over 150 customers that are currently distributing its data under different integration models. ● Analyzing over 50TB of data ● Leveraging machine learning to derive actionable insights about crypto assets FINTECH AND NEWS SITES (45) WALLETS (5) DATA PROVIDERS (9) BLOCKCHAIN PROJECTS (5) EXCHANGES (35)
  • 3.
    Quant Intelligent ServicesUnit Traction ● ITB has partnered with the top quant funds in the industry to provide alpha- generating intelligence. ● Over $500M deployed in ITB crypto quant strategies ● ITB receives a % of the profits generated by each quant fund’s strategy that uses ITB services. QUANT IN CRYPTO LAND
  • 4.
    Agenda ● Unique characteristicsof quant strategies in crypto ● Unique sources of alpha ● DeFi as a new frontier for HFTs ● Inefficiencies in crypto ● NFTs as a new form of asset class ● Why most crypto quant strategies fail? ● Cutting edge deep learning and crypto quant strategies QUANT IN CRYPTO LAND
  • 5.
    Some factors thatmakes crypto unique for quant strategies...
  • 6.
    QUANT IN CRYPTOLAND Unique Sources of Alpha New Programmable Financial Primitives Regularly Inefficient Market Four Core Factors New Types of Digital Assets 1 2 3 4
  • 7.
    Unique Sources ofAlpha… QUANT IN CRYPTO LAND
  • 8.
    Blockchain datasets representa very unique dataset to understand the behavior of groups of investors… QUANT IN CRYPTO LAND
  • 9.
  • 10.
    Support and ResistanceBased on Wallet Holdings Ex: Bitcoin QUANT IN CRYPTO LAND
  • 11.
    Event Signals Basedon Flows Into Exchanges Ex: Bitcoin QUANT IN CRYPTO LAND
  • 12.
  • 13.
    Patterns About HoldingTime, Geography of Relationships of Individual Investor Positions Ex: Bitcoin QUANT IN CRYPTO LAND
  • 14.
    And many more... QUANTIN CRYPTO LAND
  • 15.
    Blockchain datasets canprovide a unique set of features to quant models that can lead to robust alpha capture… QUANT IN CRYPTO LAND
  • 16.
  • 17.
    QUANT IN CRYPTOLAND Decentralized finance(DeFi) enables the representations of financial primitives as decentralized programmable smart contract… Lending Market Making Capital Formation Asset Composition Insurance Derivatives
  • 18.
    Programmable, on-chain, financialprimitives represent a new frontier for high frequency trading… QUANT IN CRYPTO LAND
  • 19.
    The DeFi HFTPerspective QUANT IN CRYPTO LAND Block Time Speed Factor Trade Transparency Factor Cost Factor MEV Factor
  • 20.
    The DeFi HFTPerspective QUANT IN CRYPTO LAND Block Time Speed Factor Trade Transparency Factor Cost Factor MEV Factor The HFT DeFi Transactions are limited by the processing time of each block The Result Speed advantage is relatively limited
  • 21.
    The DeFi HFTPerspective QUANT IN CRYPTO LAND Block Time Speed Factor Trade Transparency Factor Cost Factor MEV Factor The HFT DeFi All trades are visible to all competing parties The Result Every single HFT trade can be copied or subject to competition
  • 22.
    The DeFi HFTPerspective QUANT IN CRYPTO LAND Block Time Speed Factor Trade Transparency Factor Cost Factor MEV Factor The HFT DeFi There is a cost associated with transaction execution The Result Cost becomes an economic factor evaluating the viability of specific trades
  • 23.
    The DeFi HFTPerspective QUANT IN CRYPTO LAND Block Time Speed Factor Trade Transparency Factor Cost Factor MEV Factor The HFT DeFi Trades are dependent on miner’s criteria for placing transactions in a block The Result Perfectly viable HFT trades can fail because of the mining criteria
  • 24.
  • 25.
    QUANT IN CRYPTOLAND Information Asymmetry Insiders in crypto projects have regular access to material, market influencing information News Overreaction News in crypto cause dispositional movements in the market Massive Event Driven Trades New token listings, project updates… Crypto Market Inefficiencies CeFi-DeFi Arbitrages Asset prices in DeFi and CeFi exchanges are regularly influenced by different factors Cross Protocol Arbitrages Lending, borrowing, market making dynamics are drastically different across different protocols
  • 26.
  • 27.
    Crazy Correlations withTwitter Sentiment Ex: Cardano QUANT IN CRYPTO LAND
  • 28.
    DeFi Borrow-Lending RatesArbitrage QUANT IN CRYPTO LAND
  • 29.
    New types ofdigital assets… QUANT IN CRYPTO LAND
  • 30.
    Non-Fungible tokens area typical case of a crypto-based asset class that trades based on different dynamics than the underlying platform QUANT IN CRYPTO LAND
  • 31.
  • 32.
  • 33.
  • 34.
    Challenges of CryptoQuant Strategies… QUANT IN CRYPTO LAND
  • 35.
    Most crypto quantstrategies run into very peculiar challenges… QUANT IN CRYPTO LAND
  • 36.
  • 37.
    QUANT IN CRYPTOLAND Datasets are really small compared to traditional capital markets Crypto is Different Most ML quant models struggle learning with small datasets The availability of existing high quality datasets is very unstable
  • 38.
    Challenge #2 Regular “outlier”events QUANT IN CRYPTO LAND
  • 39.
    QUANT IN CRYPTOLAND Outlier events happen regularly(ex: Bitcoin recent rally, March crash, the raise of DeFi…) Crypto is Different ML quant models struggle to learn patterns they haven’t seen in training datasets Some outlier events introduce noise in training datasets
  • 40.
    Challenge #3 The regularretraining dilemma QUANT IN CRYPTO LAND
  • 41.
    QUANT IN CRYPTOLAND Outlier events cause ML crypto quant strategies to fail and force them to incorporate new records in the training dataset Crypto is Different Training ML crypto quant models with the latest data is a recipe for overfitting Outlier training records could distort the behavior of crypto quant strategies
  • 42.
    Challenge #4 Data qualityand reliability QUANT IN CRYPTO LAND
  • 43.
    QUANT IN CRYPTOLAND Crypto datasets are plagued with “bad data”(ex: wash trades, fake volume…) Crypto is Different Crypto data providers are unstable and fail all the time ML crypto quant models fail with bad or missing data
  • 44.
    Challenge #5 Anonymous blockchainrecords QUANT IN CRYPTO LAND
  • 45.
    QUANT IN CRYPTOLAND Anonymous blockchain records introduce a misleading level of uniformity in training datasets in ML crypto quant models Crypto is Different There is no consistency in the way crypto exchanges interact with the underlying blockchains There is no linear correlation between blockchain activity and trading activity
  • 46.
    Challenge #6 Simple modelfallacy QUANT IN CRYPTO LAND
  • 47.
    QUANT IN CRYPTOLAND Crypto is too inefficient for simple ML quant strategies Crypto is Different It’s hard to model the behavior of crypto assets with a handful of parameters Quant research based on simple ML models results impractical in the crypto space
  • 48.
    Some cutting-edge ideasthat might influence quant strategies in crypto… QUANT IN CRYPTO LAND
  • 49.
    Crypto is coincidingwith the renaissance of deep learning methods which can have a disproportional impact in quant strategies in the space… QUANT IN CRYPTO LAND
  • 50.
    QUANT IN CRYPTOLAND Graph Neural Networks Blockchain data is fundamentally hierarchical and well suited for GNN models Transformers for Time Series Traditional time series forecasting models will struggle with the complexities of the crypto market. Transformer models applied to time series datasets can have strong predictive capabilities Self-Supervised Learning The anonymity of blockchain datasets it’s a challenge for most supervised learning methods. Self-supervised learning techniques can organically learn patterns using unlabeled blockchain datasets Crypto Market Inefficiencies Reinforcement Learning for On- Chain Data The on-chain transparency of DeFi trades makes it a good candidate for reinforcement learning models
  • 51.
    Some practical lessonsabout building quant strategies in crypto… QUANT IN CRYPTO LAND
  • 52.
    QUANT IN CRYPTOLAND The Simplest Trades Work Market inefficiencies cause very simple trading strategies to work in crypto Ex: basis trading. Assume Very Fragile Quant Infrastructures Poor quality datasets, regular exchange failures. Traditional Quant Strategies can be Adapted to Crypto Many traditional order book or derivative strategies can be retrained for crypto assets. Crypto Market Inefficiencies Plenty of Opportunities for New Strategies DeFi, new derivatives open the door to new types of strategies.
  • 53.
    Summary ● Crypto representsa new landscape for quant strategies ● The immaturity of the crypto space creates very unique challenges for quant strategies ● DeFi represents a new frontier for HFT quant strategies ● NFTs are a great example of new trading dynamics ● New deep learning methods could have an impact in crypto quant strategies QUANT IN CRYPTO LAND
  • 54.