Inventory prediction in foreign exchange markets Damien Challet - PowerPoint PPT Presentation

Inventory prediction in foreign exchange markets Damien Challet CentraleSupélec and Encelade Capital with Mehdi Lallouache, Rémy Chicheportiche, Serge Kassibrakis Mosaic Finance, Capital Fund Management, Swissquote Bank May 12, 2017 with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Near-prejudices in Finance Public data: price nearly unpredictable Proprietary data: nearly unaccessible, underexploited Machine learning: nearly useless, too much noise with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Foreign exchange market structure Exchange EUR → USD 10,000 Liquidity Liquidity 100,000 Provider Provider BROKER -30,000 Liquidity Provider 1,000,000 with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Broker inventory Minimum amount to Liquidity Provider 10,000 Liquidity Liquidity 100,000 Broker aggregates small Provider Provider amounts BROKER → inventory -30,000 Liquidity Provider 1,000,000 Order matching: save exchange fees with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Inventory management Φ MAX ∈ [ 100 , 000 , 10 , 000 , 000 ] Set maximum inventory φ t ∼ P ( φ ) Random client order flow Φ t = ∑ t ′ ≤ t φ t Keep orders in inventory if | Φ t | > Φ MAX Flush inventory to liquidity provider dP t Random price changes P t = trend + noise t Mathematical Finance: optimise risk as a function of Φ MAX and time horizon T with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Big data approach to inventory management Number of transactions per client 10 0 10 -1 O ( 10 4 ) clients P > ( N ) 10 -2 O ( 10 7 ) transactions 10 -3 10 0 10 1 10 2 10 3 10 4 N Financial markets are predictable with client IDs with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Prediction problems = Science + Engineering Science: flux cluster clients 1 lead-lag networks 2 machine learning 3 ENJOY! 4 Engineering Predict price returns 1 Inventory constraints 2 PROFIT! 3 with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Cluster traders ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ●● ● 2 ● ● ● ● ● ● 10 ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● ● ● ● ● ● ● 5 ● ● ● ● ● ● ● ● ● ● cumsum(strat) cumsum(strat) ● ● −2 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● 0 ● ● ● ● −4 ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● −5 −6 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● −8 ● ● ● ● ●● ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ●● ● −10 ● ● ● ● ● ● ● ● ● ● 50 100 150 200 50 100 150 200 yday yday 50 8 ● ●● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● ● ● ● ● ● ● 40 ● ● ● ● ● ● 6 ● ● ● ● ●● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● 30 ● ● ● ● cumsum(strat) 4 ● ● ● ● ● ●● ● ●● ● ● ● cumsum(strat) ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● ● ●● 20 ● ● ● ● ● ● ●● ● ● ● ● ● ● 2 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 10 ● ● ● ● ● ●● ● ● ● ● ●● ● ●● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● 0 ● ● ● ● ● ●● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ●● ● ● ● ● ● −2 ● ● ● ● ● 50 100 150 200 50 100 150 200 yday yday with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Unsupervised classification: synchronous actions Tumminello et al. (2011a) Agent i , state i ( t ) ∈ { 1 , ··· , S } 1 s i ( t ) = sign Buy ( t ) − Sell ( t ) Buy ( t )+ Sell ( t ) ∈ {− 1 , 0 , 1 , 2 = / 0 } H 0 : states of agents i and j are Poisson processes O ( N 2 ) pairs → multiple hypothesis correction → NETWORK with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Synchronized traders Challet et al. (2017) EUR.USD SQ daily hourly Links Links ● ● ● ● ● −1 −1 ● ● ● −1 −1 ● ● ● ● ● ● ● ● ● 1 1 ● ● ● 1 1 ● ● ● ● ● 2 2 ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● ● with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets

Predictability of order flow Conditions: Clusters of agents are stable 1 Cluster activity is predictable 2 ≡ persistent lead-lag cluster networks with Mehdi Lallouache, Rémy Chicheportiche, Damien Challet CentraleSupélec and Encelade Capital Inventory prediction in foreign exchange markets