Generative Adversarial Networks for Amplifying and Extending - - PowerPoint PPT Presentation

▶
generative adversarial networks for amplifying and
SMART_READER_LITE
LIVE PREVIEW

Generative Adversarial Networks for Amplifying and Extending - - PowerPoint PPT Presentation

Jun 07, 2023 408 likes •655 views

Generative Adversarial Networks for Amplifying and Extending Financial Market Data Michael Wellman Lynn A. Conway Collegiate Professor of Computer Science & Engineering University of Michigan References Generating realistic stock market

slide-1
SLIDE 1

Generative Adversarial Networks for Amplifying and Extending Financial Market Data

Michael Wellman

Lynn A. Conway Collegiate Professor of Computer Science & Engineering University of Michigan

slide-2
SLIDE 2

References

  • Generating realistic stock market order streams

(J. Li, X. Wang, Y. Lin, A. Sinha, and M. P. Wellman). 34th AAAI Conference on Artificial Intelligence, pages 727–734, Feb 2020.

  • Market manipulation: An adversarial learning

framework for detection and evasion (X. Wang and

  • M. P. Wellman). 29th International Joint Conference
  • n Artificial Intelligence, pages 4626–4632, July

2020.

slide-3
SLIDE 3

Spoofing is the practice of submitting large spurious buy or sell orders with the intent to cancel them before execution to mislead other traders.

3

slide-4
SLIDE 4

Source: Financial Conduct Authority, Animated Example of Mr. Coscia’s Trading

18

115.90 - 115.89 - 115.88 - 115.87 - 115.86 - 115.85 - 115.84 - 115.83 - 115.82 - 115.81 - True buy order True sell order Profit ms 100 200 300 400 500 600 Price Transacted sell Transacted buy Large spoof buy orders Large spoof sell orders

slide-5
SLIDE 5

Detecting Market Manipulation

  • The ideal case: adopt supervised learning approaches
  • Represent an order stream as a variable-length sequence of bidding actions

(e.g., price and quantity pairs)

19

Order Streams from Individual Traders Manipulation or Normal Trading

Detector

slide-6
SLIDE 6

Developing Manipulation Signatures for Detection

  • Given model of spoofing behavior, need calibrated market data

source for injection

  • Approach: Learn to generate realistic financial order streams

background trading model

market data

calibration

spoofing strategy

  • ptimization

signature extractor

surveillance/audit algorithms

machine learning

slide-7
SLIDE 7

Order Book Evolution

  • Generator outputs next order, conditional on order

book state and history

slide-8
SLIDE 8

Generative Adversarial Networks (GANs)

Interleave training of two deep NNs: Generator

  • Takes noise vector as input, generates sample data item
  • bjective: confuse critic

Critic

  • Takes real or generated sample, classifies as real or not
slide-9
SLIDE 9

Generator NN Architecture

LSTM layer Noise input Convolutional layers after a single fully connected layer Pre-trained CDA network Time History xj of length k

slide-10
SLIDE 10

Critic NN Architecture

LSTM layer Convolutional layers after a single fully connected layer Output Time

History xj of length k Real/Gen xi

slide-11
SLIDE 11

Experiments and Evaluation

  • Trained on:
  • 1. simulated financial market
  • 2. thinly traded stock: PN, 20K orders/day
  • 3. thickly traded stock: GOOG, 230K orders/day
  • Evaluated various statistics comparing real and generated (fake) order

streams

slide-12
SLIDE 12

Results (Buy Orders)

slide-13
SLIDE 13

Results (Bid/Ask Evolution)

slide-14
SLIDE 14

Detecting Manipulation: Challenges

  • Codified manipulation strategies may not be diverse enough;
  • Adversary may obfuscate actions to evade detection, given a

developed classifier.

28

slide-15
SLIDE 15

29

An Adversarial Learning Framework to Evade Detection

  • A case study: modify spoofing to resemble market making.
  • A market maker provides liquidity by simultaneously submitting buy and sell
  • rders around an estimate of the fundamental value.

MM SP

D0

<latexit sha1_base64="Qi6CBnHLXm9RFUR7UBemC2sIbQ=">AB9HicbVDLSgMxFL3js9ZX1aWbYBFclRkp6LKoC5cV7APaoWTStA3NZMbkTqEM/Q43LhRx68e482/MtLPQ1gOBwzn3ck9OEth0HW/nbX1jc2t7cJOcXdv/+CwdHTcNFGiGW+wSEa6HVDpVC8gQIlb8ea0zCQvBWMbzO/NeHaiEg94jTmfkiHSgwEo2glvxtSHDEq07tZz+2Vym7FnYOsEi8nZchR75W+uv2IJSFXyCQ1puO5Mfop1SiY5LNiNzE8pmxMh7xjqaIhN346Dz0j51bpk0Gk7VNI5urvjZSGxkzDwE5mIc2yl4n/eZ0EB9d+KlScIFdscWiQSIRyRogfaE5Qzm1hDItbFbCRlRThranoi3BW/7yKmleVrxqpfpQLdu8joKcApncAEeXEN7qEODWDwBM/wCm/OxHlx3p2Pxeiak+cwB84nz+jI5IE</latexit>

1000 2000 3000 4000 5000 TiPe 99000 99400 99800 100200 100600 101000 3rice 1000 2000 3000 4000 5000 TiPe 99000 99400 99800 100200 100600 101000 3rice

A Manipulation Order Stream (SP) A Market-Making Order Stream (MM)

slide-16
SLIDE 16

32

An Adversarial Learning Framework to Evade Detection

SP MM Market Simulator SP SP1 Market Simulator

Lreg

<latexit sha1_base64="xf5t+RLxbFtdzKb5ZJ5kyMdOZmc=">AB/3icbVBNS8NAFNzUr1q/oIXL8EieCqJFOyx4MWDhwq2FZpSNtvXdulmE3ZfxBJ78K948aCIV/+GN/+NmzYHbR1YGbe481OEAu0XW/rcLK6tr6RnGztLW9s7tn7x+0dJQoBk0WiUjdBVSD4BKayFHAXayAhoGAdjC+zPz2PSjNI3mLkxi6IR1KPuCMopF69pEfUhwxKtLrac9HeMBUwXDas8tuxZ3BWSZeTsokR6Nnf/n9iCUhSGSCat3x3Bi7KVXImYBpyU80xJSN6RA6hkoagu6ms/xT59QofWcQKfMkOjP190ZKQ60nYWAms7R60cvE/7xOgoNaN+UyThAkmx8aJMLByMnKcPpcAUMxMYQyxU1Wh42ogxNZSVTgrf45WXSOq941Ur1plqu1/I6iuSYnJAz4pELUidXpEGahJFH8kxeyZv1ZL1Y79bHfLRg5TuH5A+szx/8Zpa0</latexit>

Ladv

<latexit sha1_base64="iHEBTNGWZadnMDkzl0OKw5qR8=">AB/3icbVDLSsNAFJ3UV62vqODGTbAIrkoiBbsuHhoJ9QBPCZDJph04ezNwUS8zCX3HjQhG3/oY7/8ZJm4W2Hhg4nHMv98zxEs4kmOa3Vlb39jcqm7Xdnb39g/0w6OejFNBaJfEPBYD0vKWUS7wIDTQSIoDj1O+97kuvD7Uyoki6N7mCXUCfEoYgEjGJTk6id2iGFM9uc9cG+gAZ9qe5q9fNhjmHsUqsktRiY6rf9l+TNKQRkA4lnJomQk4GRbACKd5zU4lTCZ4BEdKhrhkEonm+fPjXOl+EYQC/UiMObq740Mh1LOQk9NFmnlsleI/3nDFIKWk7EoSYFGZHEoSLkBsVGUYfhMUAJ8pgmgqmsBhljgQmoymqBGv5y6ukd9mwmo3mXbPebpV1VNEpOkMXyEJXqI1uUAd1EUGP6Bm9ojftSXvR3rWPxWhFK3eO0R9onz/3tJax</latexit>
  • Manipulation intensity

the fraction of price deviation;

  • Transaction risk

# transactions / # arrivals

D0

<latexit sha1_base64="p3I1zsXs0PIZIi2DkU+U0KLK7TQ=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkoMeiHjxWtB/QhrLZTtqlm03Y3Qgl9Cd48aCIV3+RN/+N2zYHbX0w8Hhvhpl5QSK4Nq7RTW1jc2t4rbpZ3dvf2D8uFRS8epYthksYhVJ6AaBZfYNwI7CQKaRQIbAfjm5nfkKleSwfzSRBP6JDyUPOqLHSw23f7ZcrbtWdg6wSLycVyNHol796g5ilEUrDBNW67mJ8TOqDGcCp6VeqjGhbEyH2LVU0gi1n81PnZIzqwxIGCtb0pC5+nsio5HWkyiwnRE1I73szcT/vG5qwis/4zJDUq2WBSmgpiYzP4mA6QGTGxhDLF7a2EjaizNh0SjYEb/nlVdK6qHq1au2+Vqlf53EU4QRO4Rw8uIQ63EDmsBgCM/wCm+OcF6cd+dj0Vpw8plj+APn8we9K41y</latexit>

G1

<latexit sha1_base64="Yr1pBXtGQ0+Rtn2nqEmST9IduXY=">AB6nicbVBNS8NAEJ3Ur1q/qh69LBbBU0mkoMeiBz1WtB/QhrLZTtqlm03Y3Qgl9Cd48aCIV3+RN/+N2zYHbX0w8Hhvhpl5QSK4Nq7RTW1jc2t4rbpZ3dvf2D8uFRS8epYthksYhVJ6AaBZfYNwI7CQKaRQIbAfjm5nfkKleSwfzSRBP6JDyUPOqLHSw23f65crbtWdg6wSLycVyNHol796g5ilEUrDBNW67mJ8TOqDGcCp6VeqjGhbEyH2LVU0gi1n81PnZIzqwxIGCtb0pC5+nsio5HWkyiwnRE1I73szcT/vG5qwis/4zJDUq2WBSmgpiYzP4mA6QGTGxhDLF7a2EjaizNh0SjYEb/nlVdK6qHq1au2+Vqlf53EU4QRO4Rw8uIQ63EDmsBgCM/wCm+OcF6cd+dj0Vpw8plj+APn8wfDQY12</latexit>
  • Adapt SP to evade detection while preserving manipulation effects
slide-17
SLIDE 17

An Adversarial Learning Framework to Evade Detection

33

  • A recursive training procedure
slide-18
SLIDE 18

Empirical Evaluation

34

  • Similarity to market making;
  • Preservation of manipulation effects.
slide-19
SLIDE 19

Similarity to Market Making

35

Ø Quote simultaneously on both sides of the market; Ø Place large orders behind smaller ones.

SP SP1 SP SP2 SP SP3 SP SP

slide-20
SLIDE 20

SP1 → SP2 → SP3 →

  • Closer resemblance to MM
  • Orders cover a wider range of prices with small quantities
  • Buy and sell orders are better balanced
  • Degraded spoofing effect
  • reduced manipulation intensity
  • higher transaction risk
slide-21
SLIDE 21

Combine Agent-Based Simulation and Adversarial Learning to Detect Market Manipulation

37

Generate New Manipulation Patterns Verify the Manipulation Effects

slide-22
SLIDE 22

Recap: GANs for Financial Modeling

  • Train model to generate realistic order streams
  • An adversarial learning framework for market manipulation
  • Reason about how a manipulator might mask its behavior
  • Understand the dynamics of evasion and detection
  • Generate a diverse set of manipulative patterns to improve detection

robustness

38