1. Market Science: How markets could be studied as a hard science Edward Tsang CCFEA University of Essex ColchesterColchester, UK Colchester 21 August 2015

2. Studying Financial Markets 21 August 2015 Agent-based studies: Simulate markets Look for stylised facts Artificial Market EDDIE Intelligence EDDIE Fundamental EDDIE Noise Technical Analysis: Discover regularities by analysing past series Classical economics: Mathematical analysis Find fundamental values Market Science: Observe micro-behaviour Discover regularities

3. How Financial Markets are studied 21 August 2015 Technical Analysis: Discover regularities by analysing past series Classical economics: Mathematical analysis Find fundamental values

4. How to Understand Financial Markets? Edward Tsang CCFEACCFEA / CIC CIC CCFEACIC University of Essex ColchesterColchester, UK Colchester 21 August 2015

5. Classical Economics To model economy and prices mathematically To model economy and prices mathematically Built on critical assumptions Built on critical assumptions Perfect rationality Perfect rationality Perfect rationality Perfect rationality Homogeneity Homogeneity Homogeneity Market efficiency Market efficiency Market efficiency Market efficiency But market has changed, e.g. But market has changed, e.g.market has changedmarket has changed Increased risk-taking Increased risk-taking High frequency trading High frequency trading Leading to the butterfly effect! Leading to the butterfly effect!butterfly effectbutterfly effect A small, ordinary event… A small, ordinary event… … could lead to avalanche … could lead to avalanche 21 August 2015

6. Classical Economics 21 August 2015 Built on critical assumptions Everybody is perfectly rational Everybody thinks the same Everybody has full information Market has changed! Increased risk-taking Computer trading

7. Classical Economics To model economy and prices mathematically To model economy and prices mathematically Classic: Capial Asset Pricing Model (CAPM) Classic: Capial Asset Pricing Model (CAPM) E(R i ) = R f + β im (E(R m ) - R f ) where β im = Cov(R i, R m ) / Var(R m ) E(R i ) is the expected return on the capital asset i E(R i ) is the expected return on the capital asset i R f is the risk-free rate of interest R f is the risk-free rate of interest β im is the sensitivity of the asset returns to market returns β im is the sensitivity of the asset returns to market returns E(R m ) is the expected return of the market E(R m ) is the expected return of the market Built on important assumptions Built on important assumptions e.g. perfect rationality, market efficiency, homogeneity e.g. perfect rationality, market efficiency, homogeneityperfect rationalityhomogeneityperfect rationalityhomogeneity Market has changed, leading to the butterfly effect! Market has changed, leading to the butterfly effect! Market has changedbutterfly effect Market has changedbutterfly effect 21 August 2015

8. Changes in the market Computer trading Computer trading Programs react in milliseconds Programs react in milliseconds New financial instruments New financial instruments E.g. Options, CFDs E.g. Options, CFDs Increased leverage Increased leverage More people can influence the market More people can influence the market More data available More data available Both historical data and transaction data Both historical data and transaction data Computational methods developed Computational methods developed Machine learning more efficient and effective Machine learning more efficient and effective 21 August 2015

9. The butterfly effect Movements are amplified Movements are amplified By increased leverage By increased leverage By computer trading By computer trading Cascading margin calls Cascading margin calls Leading to avalanche Leading to avalanche Can it be studied? Can it be studied? Yes, if one looks into the microstructure! Yes, if one looks into the microstructure!microstructure 21 August 2015

10. Technical Analysis (Chartists) 21 August 2015 Attempt to find patterns in the chart in order to predict future movements Patterns have been found. People made money with them. Questions: Do these patterns repeat? How often? As traders’ behaviour change, old patterns become irrelevant. E.g. Algorithmic trading replaces human trading

11. Agent-based Market Studies Attempt to model trading agents Attempt to model trading agents Simulate their interaction Simulate their interaction Contrast their behaviour against real markets Contrast their behaviour against real markets If stylised facts can be observed, then the model could be useful for finding plausible scenarios If stylised facts can be observed, then the model could be useful for finding plausible scenarios Doesn’t benefit forecasting (see chaos theory) Doesn’t benefit forecasting (see chaos theory) But useful for understanding But useful for understanding Also useful for market design / policies Also useful for market design / policies 21 August 2015 CHASM, an artificial market platform

12. 21 August 2015 All rights reserved, Edward Tsang 21 August 2015 Agent-based Markets Studies Artificial Market Agent 1 Agent 2 Agent n Experimenter E.g. supply prices Prices, wealth, etc 4. Compare and contrast 2. interaction Observed market data (“stylized facts”) 3. Observe 5. modify 1. modelling CHASM, an artificial market platform “All models are wrong, but some are useful” (Box 1987) Why?Where applied?

13. Why Model-based Markets? Possible futures Real data only shows one history Simulation could reveal possible crises One could block the paths to crises Establish causal relations Change agent bahaviour or market conditions Observe differences Differences are due to controlled changes 21 August 2015

14. Artificial Market Applications Find conditions under which crashes happen Build early warning system for financial crisis Find consequences of bank failure Implication on regulations Find subgame equilibrium Help deciding how to set the rules or play the game Test algorithmic trading strategies Wind-tunnel testing before deployment 21 August 2015

15. The “Biology” of Markets (Richard Olsen) How was biology studied? How was biology studied? 21 August 2015 Richard Olsen Forex OANDA Observe Observe Copy Copy Measure Measure Generalize Generalize …

16. Market Studied as Hard Science Markets are results of micro-behaviour Markets are results of micro-behaviour Technical analysis only studies the results (prices) Technical analysis only studies the results (prices) Much deeper knowledge can be observed from studying micro-behaviour … Much deeper knowledge can be observed from studying micro-behaviour … We may not know who’s going to buy/sell We may not know who’s going to buy/sell But we can study the state of the market But we can study the state of the market And ask what-if questions And ask what-if questions Or questions about its liquidity Or questions about its liquidity 21 August 2015

17. Analogy with weather forecast Weather forecast: Weather forecast: We may not know the long term weather changes We may not know the long term weather changes We have lots of sensor information We have lots of sensor information We know air flows from high to low pressure regions We know air flows from high to low pressure regions We can predict the short-term weather with some confidence We can predict the short-term weather with some confidence Market Science Market Science We know exactly what orders are placed We know exactly what orders are placed We know how orders are cleared We know how orders are cleared We know exactly what happens next We know exactly what happens next 21 August 2015

18. Data-driven Market Studies If I can model every investor, I can predict the market If I can model every investor, I can predict the market But I can’t accurately model investors… But I can’t accurately model investors… However, I know exactly what orders were placed However, I know exactly what orders were placedexactly what orders were placedexactly what orders were placed I know what happened after each order was placed I know what happened after each order was placedwhat happened after each order was placedwhat happened after each order was placed Whether it was transacted Whether it was transacted Its immediate impact to prices Its immediate impact to prices Price movements afterwards Price movements afterwards Can’t we learn anything from these? Can’t we learn anything from these? By recording details and looking for regularities By recording details and looking for regularities As we do in biology As we do in biology 21 August 2015

19. Formal Description of Market Attempt to define market dynamics formally Attempt to define market dynamics formally to avoid ambiguity in verbal descriptions to avoid ambiguity in verbal descriptions A model tell us exactly what factors to consider A model tell us exactly what factors to consider Markets can be described by states Markets can be described by states Events change the state of the market Events change the state of the market We want to study consequences of events We want to study consequences of events Maintain consequential closure if possible Maintain consequential closure if possible How big an order would cause a crash of 20%? How big an order would cause a crash of 20%? 21 August 2015

20. Market Calculus Demo First step towards market science First step towards market science Enable scientific reasoning about markets Enable scientific reasoning about markets 21 August 2015

21. Proposal Indonesian tsunami 2004 led to construction of early warning systems Indonesian tsunami 2004 led to construction of early warning systems Economic turmoil demands the same! Economic turmoil demands the same! Banking sector lost in the crisis over US$1,000bn (£702bn) Banking sector lost in the crisis over US$1,000bn (£702bn) Investing 2%, or US$2bn is not too much Investing 2%, or US$2bn is not too much Richard Olsen, OANDA and CCFEA Richard Olsen, OANDA and CCFEA Clive Cookson, Financial Times Clive Cookson, Financial Times 21 August 2015

22. Olsen Routes – Computer-aided market analysis Routes: a programming environment for real time applications 21 August 2015

23. A Wiki-style repository of software for studying finance 21 August 2015 EconomistVisualization Expert Computational Intelligence Expert High-frequency data (including foreign exchange rates, stock and option prices, interest rates, etc) Inter- connected modules implementing models Modules interact with each other or users Users...... Users upload or retrieve modules Possibly through automated interaction Web-based Open-source

24. Concluding Summary Classical economics build castles on sand Classical economics build castles on sand Due to unrealistic assumptions Due to unrealistic assumptions Technical analysis only scratches the surface Technical analysis only scratches the surface Agent-based modelling helps understand markets Agent-based modelling helps understand markets Repeatable, enabling scientific studies Repeatable, enabling scientific studies Market science looks into micro-behaviour Market science looks into micro-behaviour Chartists look at end results, why not look at causes!? Chartists look at end results, why not look at causes!? Could be seen as a branch of behavioural finance Could be seen as a branch of behavioural finance 21 August 2015

25. Supplementary Information 21 August 2015

26. High Frequency Data: Example of an Order Book PriceVolumeOrders Seller 4 3.862,0001 Seller 3 3.8510,0005 Seller 2 3.845,0001 Seller 1 3.831,0001 Buyer 1 3.826,0003 Buyer 2 3.818,0003 Buyer 3 3.805,0001 Buyer 4 3.7917,0003 21 August 2015

27. Different investors react differently to the same piece of information The length of the coast line (profit opportunities) depends on how you measure it The length of the coast line (profit opportunities) depends on how you measure it A trader that reacts monthly (red line) has higher potential for profit than one who reacts quarterly (yellow line) A trader that reacts monthly (red line) has higher potential for profit than one who reacts quarterly (yellow line) Even with perfect foresight, one may be buying when the other is selling (April) Even with perfect foresight, one may be buying when the other is selling (April) 21 August 2015 JanAprOctJul Jan Time Price

28. Price Distribution 21 August 2015

29. Definitions of directional changes 21 August 2015

30. Striking observation When a directional change of r% occurs, it is followed by an overshoot of r% When a directional change of r% occurs, it is followed by an overshoot of r% Observing the power law Observing the power law The time for the overshoot to happen is also highly correlated to the time taken for the change of direction to happen! The time for the overshoot to happen is also highly correlated to the time taken for the change of direction to happen! Further observation and analysis needed Further observation and analysis needed 21 August 2015

31. Length of coastline 21 August 2015 Maximum profit opportunity after transaction costs with no leverage and perfect foresight Long coast line (>2,000%) means huge opportunities to be exploited! opportunities

32. Observation from OANDA Plotting Long/Short against Win/Lost positions Plotting Long/Short against Win/Lost positions On EUR/USD on 31 st March 2008 On EUR/USD on 31 st March 2008 There are more losers than winners There are more losers than winners There are more short than long positions There are more short than long positions Rich information to be analysed (PhD projects) Rich information to be analysed (PhD projects) 21 August 2015 Those who bought Euro with USD at prices higher than current price (i.e. in potential loss) Green: in profit Blue: in loss Those who short Euros for USD at prices higher than current price (i.e. with unrealized profit) Long Short Current price

33. Market Analysis You can play Forex game at OANDA You can play Forex game at OANDA 21 August 2015

34. Vernon Smith Nobel Prize in Economics 2002 Nobel Prize in Economics 2002 Experimental Economics Experimental Economics “for having established laboratory experiments as a tool in empirical economic analysis, especially in the study of alternative market mechanisms” George Mason University, USA George Mason University, USA 21 August 2015

35. Agent-based Artificial Markets Better understand the market Better understand the market What makes a market efficient? What makes a market efficient? Phenomenological approach Phenomenological approach 21 August 2015 Artificial Market Agent 1 Agent 2 Agent n What happens when agents evolve? Nash equilibrium Wind Tunnel Market Testing Designing new markets Strategy Design How to do well in market Fundamental Application exogenous endogenous 內在自決 外部控制 人工市场 人工代理

36. Summary: Agent-based Artificial Markets Strategies evolved Strategies evolved Payments market Payments market Payments market Payments market Bargaining Bargaining Bargaining Wind-tunnel Testing British Telecom British Telecom Road usage market Road usage market 21 August 2015 Nash Equilibrium Nash Equilibrium Nash Equilibrium Nash Equilibrium Red Queens effect studied Red Queens effect studied Reproduced stylized facts in CHASM Reproduced stylized facts in CHASMCHASM Rationality challenged Rationality challenged Rationality Fundamental Applications AgentsMarkets

37. Micro behaviour analysis Approach: Approach: Modelling trading agents Modelling trading agents Consequences analysis on big offers/bids Consequences analysis on big offers/bids Finding patterns, such as scaling laws Finding patterns, such as scaling laws Hope to explain market behaviour that conventional economics failed to explain Hope to explain market behaviour that conventional economics failed to explain No perfect rationality No perfect rationality No homogeneous behaviour by traders No homogeneous behaviour by traders An exciting way forward An exciting way forward 21 August 2015

38. Efficient Market Hypothesis Financial assets (e.g. shares) pricing: Financial assets (e.g. shares) pricing: All available information is fully reflected in current prices All available information is fully reflected in current prices If EMH holds, forecasting is impossible If EMH holds, forecasting is impossible Random walk hypothesis Random walk hypothesis Random walk hypothesis Random walk hypothesis Assumptions: Assumptions: Efficient markets (one can buy/sell quickly) Efficient markets (one can buy/sell quickly) Perfect information flow Perfect information flow Rational traders Rational traders 21 August 2015

39. Does the EMH Hold? It holds for the long term It holds for the long term But “Fat Tail ” was observation: But “Fat Tail ” was observation: big changes today often followed by big changes (either up or down) tomorrow big changes today often followed by big changes (either up or down) tomorrow How fast can one adjust asset prices given a new piece of information? How fast can one adjust asset prices given a new piece of information? Faster machines certainly help Faster machines certainly help So should faster algorithms (CIDER Theory) So should faster algorithms (CIDER Theory)CIDER TheoryCIDER Theory 21 August 2015

40. CIDER: Computational Intelligence Determines Effective Rationality (1) You have a product to sell. You have a product to sell. One customer offers £10 One customer offers £10 Another offers £20 Another offers £20 Who should you sell to? Who should you sell to? Obvious choice for a rational seller Obvious choice for a rational seller 21 August 2015

41. CIDER: Computational Intelligence Determines Effective Rationality (2) You are offered two choices: You are offered two choices: to pay £100 now, or to pay £100 now, or to pay £10 per month for 12 months to pay £10 per month for 12 months Given cost of capital, and basic mathematical training Given cost of capital, and basic mathematical training Not a difficult choice Not a difficult choice 21 August 2015 …

42. CIDER: Computational Intelligence Determines Effective Rationality (3) Task: Task: You need to visit 50 customers. You need to visit 50 customers. You want to minimize travelling cost. You want to minimize travelling cost. Customers have different time availability. Customers have different time availability. In what order should you visit them? In what order should you visit them? 21 August 2015  This is a very hard problem  Some could make wiser decisions than others

43. The CIDER Theory Rationality involves Computation Rationality involves Computation Rationality involves Computation Rationality involves Computation Computation has limits Computation has limits Computation has limits Computation has limits Herbert Simon: Bounded Rationality Herbert Simon: Bounded Rationality Herbert SimonBounded Rationality Herbert SimonBounded Rationality Rubinstein: model bounded rationality by explicitly specifying decision making procedures Rubinstein: model bounded rationality by explicitly specifying decision making procedures Rubinstein Decision procedures involves algorithms + heuristics Decision procedures involves algorithms + heuristics Computational intelligence determines effective rationality Computational intelligence determines effective rationality Where do decision procedures come from? Where do decision procedures come from? Designed? Evolved? Designed? Evolved? 21 August 2015

44. Top Down vs Bottom Up Approaches Top Down: write down general rules, and use them to deduce market behaviour Top Down: write down general rules, and use them to deduce market behaviour Classical economics Classical economics Agent-based approach Agent-based approach Bottom Up: observe detailed changes, and use them to induce general rules Bottom Up: observe detailed changes, and use them to induce general rules Technical analysis Technical analysis Market science Market science 21 August 2015