Transcription
THE MACHINE DESIGNOFTRADING SYSTEMS5/6/2015
YOUR PRESENTER:MIKE BARNA Founder and President, Trading System LabSr.VP Regency Stocks and Commodities Fund, LP,LLC, QEP, CPO, CTABS Mathematics, Arizona State UniversityMS Astronautical and Aeronautical Engineering, Stanford UniversitySystems Authored or Co-Authored: TSL MACHINE GENERATEDSYSTEMS, R-MESA, BIGBLUE, MESA BONDS, MESA NOTES, SIERRAHOTELFormer Defense Industry Rocket-Ramjet, Laser and Guidance EngineerStar Wars Research and Development Management EngineerSeries 3, Series 3012 FAA pilot certificates or ratingsContact: lab.com
OUR TEAM Mike Barna: Trading System Lab-Silicon Valley Based trading researchand development company with a team of international and domesticprogrammers, third party developers and testers. Developed the FirstCommercially available Machine Designed Trading Systems Platformthat requires no programming from the user.www.tradingsystemlab.com Frank Francone: Register Machine Learning, Inc.-US Based companywith a team of international and domestic machine learning scientists,IP attorneys, statisticians and programmers. Involved in governmentcontracts. Produces the LAIMGP licensed exclusively to TSL. Authoredthe leading University Textbook on GP. 1600 citations.www.rmltech.com5/6/2015
WHAT IS TSL? TSL is a platform for the Machine Design of Trading Strategies Linear Automatic Induction of Machine Code with GeneticProgramming (LAIMGP) (algorithms are nonlinear) Code is exported in different languages Strategies are tested OOS “during” design Patented and Trademarked Single Market Systems: HFT, MFT, LFT Daytrading Pairs Portfolios Options5/6/2015
OUR PRODUCT:The TSL Platform Unlimited orthogonal return streamsUse any data: No Programming RequiredAny time frameVery fastCode is exported in different languagesAnti-curve fitted and pre-tested OOS “during” designAny trading tactic: Pairs, Portfolios, Options, Daytrading, HFT5/6/2015
TSL CLIENTS AND TRADERSTSL’s JOB IS TO PROVIDE TSL TO CLIENT TRADERS Major Wall Street Investment Bank 100MSmall and Mid size CTA’s: 10M- 100MProprietary Trading Firms: 5M- 50MIndividual Traders 5MInternational Traders and FundsStrategy Development EngineersBeginner to PhD5/6/2015
REQUIRED DISCLAIMERHYPOTHETICAL PERFORMANCE RESULTS HAVE MANY INHERENT LIMITATIONS, SOME OF WHICH AREDESCRIBED BELOW. NO REPRESENTATION IS BEING MADE THAT ANY ACCOUNT WILL OR IS LIKELY TOACHIEVE PROFITS OR LOSSES SIMILAR TO THOSE SHOWN.IN FACT, THERE ARE FREQUENTLY SHARP DIFFERENCES BETWEEN HYPOTHETICAL PERFORMANCE RESULTSAND THE ACTUAL RESULTS ACHIEVED BY ANY PARTICULAR TRADING PROGRAM. ONE OF THE LIMITATIONS OFHYPOTHETICAL PERFORMANCE RESULTS IS THAT THEY ARE GENERALLY PREPARED WITH THE BENEFIT OFHINDSIGHT. IN ADDITION, HYPOTHETICAL TRADING DOES NOT INVOLVE FINANCIAL RISK, AND NOHYPOTHETICAL TRADING RECORD CAN COMPLETELY ACCOUNT FOR THE IMPACT OF FINANCIAL RISK INACTUAL TRADING. FOR EXAMPLE, THE ABILITY TO WITHSTAND LOSSES OR TO ADHERE TO A PARTICULARTRADING PROGRAM IN SPITE OF TRADING LOSSES ARE MATERIAL POINTS WHICH CAN ALSO ADVERSELYAFFECT ACTUAL TRADING RESULTS.THERE ARE NUMEROUS OTHER FACTORS RELATED TO THE MARKETS IN GENERAL OR TO THEIMPLEMENTATION OF ANY SPECIFIC TRADING PROGRAM WHICH CANNOT BE FULLY ACCOUNTED FOR IN THEPREPARATION OF HYPOTHETICAL PERFORMANCE RESULTS AND ALL OF WHICH CAN ADVERSELY AFFECTACTUAL TRADING RESULTS.5/6/2015
MACHINE LEARNING5/6/2015
MACHINES “CRUSHED”HUMAN RIVALS5/6/2015
WHAT IS A TRADING SYSTEMEQUITY CURVE?Release DateMACHINE PRODUCED ORHUMAN DESIGNED ALGORITHMS 1996 kneeS&P Futures 1982-20135/6/2015
SAMPLE COUNTERTRENDINGTRADING SYSTEMWeak Equity Curve
HUMAN DESIGNEDSTRATEGIESCOSTTIMETECHNICAL
MACHINE DESIGNEDSTRATEGIESLOWER COST PER STRATEGYREDUCED TIME TO IMPLEMENTATIONMANY TECHNICAL CAPABILITIES
BAD NEWS/GOOD NEWS Brokerage/Software companies do not or cannot provide you withthe most important item you need to be successful in the marketswith trading systems:Robust Trading Strategies that are easy to create TSL is interested only in:Robust Trading Strategies that are easy to create5/6/2015
WHAT IS THE PROBLEM TOBE SOLVED?The problem to be solved is how toconstruct a Machine that automaticallywrites better Systems faster and lessexpensively than humans can createusing manual techniques5/6/2015
SYSTEMATIC VERSES DISCRETIONARYCTA VAMI1987-2015MECHANICAL SYSTEMS457 programs, 296BSource: BarclayHedge5/6/2015DISCRETIONARY137 programs, 18B
SYSTEMATIC VERSES DISCRETIONARYCTA MUM, B 1999 to 2014350300Systematic(At least 95% Systematic)250200 B150100Discretionary(At least 65% Discretionary or Judgmental)5001999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014Source: BarclayHedge
TYPES OF SYSTEMATIC TRADINGSYSTEM DESIGN If-Then Human (manual) constructs Predictive Modeling (ML) Self-Evolving Strategy Structure (ML-TSL)1. Meta-Heuristic Simulation Based2. Supervised Learning-No Supervisory Signal3. Single and Multi-Objective4. Reinforcement LearningReference: Statistically Sound Machine Learning for Algorithmic Trading of Financial Instruments:Developing Predictive-Model-Based Trading Systems Using TSSB. David Aronson, Timothy Masters, PhD5/6/2015
PREDICTION APPROACHPREDICTIONMODELDATAEXPECTATIONNOT GOOD ENOUGH?IMPLEMENTEXPECTATION NOUGH?
PREDICTION VERSESEXPECTATION In the Prediction Approach, Prices or Volatility are forecasted n-steps out.Equity Curves are then generated as an additional step.Good Trading Systems may exhibit poor R 2Standard prediction models may require further work to generate goodEquity Curves. In the Expectation Approach, Systems are viewed as Objects and EquityCurves are generated and improved through System Metric Targeting andObjective Function Optimization.Good Trading System Equity Curves are thus directly evolved,leveraging the cross metric effect characteristic of Trading Systemsand eliminating the need for additional steps.5/6/2015
THE FAILURE OF BACKTESTS Are not proof of RobustnessHigh Potential for Over-FittingFalse sense of returnsReinforces bad design approachesLike trying to find a needle in a haystackWHAT CAN I DO ABOUT THIS ISSUE? Sequestered Data (Tests conducted in the Future) Out Of Sample Testing Walk Forward Testing Walk Backwards Testing Differential Market Testing Stress and Parametric Testing Distribution and Matched Pairs Testing(Null)Reference:” Pseudo-Mathematics and Financial Charlatanism:The Effects of Backtest Overfitting on Out-of-Sample Performance, Marcos Lopez de Prado and 3 others.5/6/2015
WHERE IS THE ALPHA?Machine Designed Trading Strategies can operate in all categoriesTRADING FREQUENCYLFTPassiveLong Holding PeriodsIndex PeggedLong Term DirectionalHedgedExpirationDerivativePortfolio DynamicsBreakoutMacro PatternGlobal MacroEvent DrivenRelative Value5/6/2015MFTShort Holding PeriodsShort Term PatternsTrend/CountertrendMean ReversionIntraday MomentumCointegrationCycle analysisVolatility BreakoutVolume analysisDaytradingSwing TradingSpreadsOptions and combinationsPortfolio of Market-ModelsHFTOrder Book PlaysSpecial Order TypesHide and LightQueue JumpingSpam and CancelTop of BookARBISOSpreadsBest ExecutionOptions surface
MACHINE BASED STRATEGYDESIGNS Operator does not need to be a programmerAllows the machine to explore a wide rangeNot limited to existing theoryWill quickly find what does and does not workMay be redirected quicklyMay be reengineered easilyHuman controlled and configured5/6/2015
TSL’s CHALLENGEIN 2007 WE CHALLENGED ANYMANUALLY DESIGNED SYSTEM TO BEATTSL’s MACHINE DESIGNED SYSTEMS INCOMPETETIVE THIRD PARTY RATINGSEVALUATED ON SEQUESTERED DATA5/6/2015
HOW IS THE SEQUESTERED DATACOMPETITION TATRAININGDATA5/6/2015OVERFITSTRATEGYWILL FAIL2008WORLDFINANCIALDISASTER
WE DID IT!In 2008, and again in 2010, TSL submitted several frozen“Machine Designed” Strategies to Futures Truth. Thesestrategies were initially held for over 18 months, then tested ately 700 submissions from over 80 worldwidestrategy designers. These systems have not been touchedsince and reporting on these “Machine Designed” Strategiescontinues through 2015.5/6/2015
THE RESULTS?MACHINE CREATED WITHNO PROGRAMMING REQUIRED2014 ReportsTSL SPon ES700 systems, 80 vendors
THE RESULTS?MACHINE CREATED WITHNO PROGRAMMING REQUIRED2015 ReportsTSL SPon ES700 systems, 80 vendors
TSL FUTURES TRUTHRATINGS OVER TIMEHighest Position all CategoriesSP, NG Systems Designed in 2007 then held for 18 monthsControlled Sequestered TestingRe-optimization not allowedSP pit closed. Systems now applied to different SymbolNote: 700 systems and 80 developer in competition5/6/2015Unfavorable BiasVariance Tradeoff(Retraining needed)
TSL MAIN COMPONENTSLEARNINGMACHINE5/6/2015
WHAT IS THE BESTLEARNING ALGORITHM?Supervised learningAODEArtificial neural network BackpropagationAutoencodersHopfield networksBoltzmann machinesRestricted Boltzmann MachinesSpiking neural networksGEGAGPLGPLAIMGP *GEPCGPGADSIFGPBayesian statistics Naive Bayes classifierBayesian networkBayesian knowledge baseCase-based reasoningDecision treesInductive logic programmingGaussian process regressionGene expression programmingGroup method of data handling (GMDH)Learning AutomataLearning Vector QuantizationLogistic Model TreeMinimum message length (decision trees, decisiongraphs, etc.)Lazy learningInstance-based learning Nearest Neighbor AlgorithmAnalogical modeling5/6/2015EMOLSKRRPCAProbably approximately correct learning (PAC)Ripple down rules, a knowledge acquisitionmethodologySymbolic machine learning algorithmsSubsymbolic machine learning algorithmsSupport vector machinesRandom ForestsEnsembles of classifiers Bootstrap aggregating(bagging)Boosting (meta-algorithm)Ordinal classificationRegression analysisInformation fuzzy networks (IFN)Conditional Random FieldStatistical classificationANOVALinear classifiers Fisher's linear discriminantLogistic regressionMultinomial logistic regressionNaive Bayes classifierPerceptronSupport vector machinesQuadratic classifiersk-nearest neighborBoostingDecision trees C4.5Random forestsBayesian networksHidden Markov modelsRef: http://en.wikipedia.org/wiki/Machine learningUnsupervised learningArtificial neural networkData clusteringExpectation-maximization algorithmSelf-organizing mapRadial basis function networkVector QuantizationGenerative topographic mapInformation bottleneck methodIBSEADAssociation rule learningApriori algorithmEclat algorithmFP-growth algorithmHierarchical clusteringSingle-linkage clusteringConceptual clusteringPartitional clusteringK-means algorithmFuzzy clusteringDBSCANReinforcement learningTemporal difference learningQ-learningLearning AutomataMonte Carlo MethodSARSADeep learningDeep belief networksDeep Boltzmann machinesDeep Convolutional neural networksDeep Recurrent neural networks
WHAT ARE THE APPLICABLESTATISTICAL TESTS? lisANOVAMedian testQ-statisticSign testFriedmanCochran Q testMcNemar testKendall coefficient of concordanceSpearman rank order RChi-squareV-square statisticPhiGammaSommer’s dPaired t-testMan-WhitneyBootstrap testCVARMonte-Carlo Permutation Tests
WHAT IS THE BEST MLSOFTWARE SUITE?AyasdiAngoss KnowledgeSTUDIOApache MahoutGesture Recognition ToolkitIBM SPSS ModelerKNIMEKXEN ModelerLIONsolverMATLABMathematicamlpyMLPACK libraryMCMLLOpenCVdlibOracle Data MiningOrangeDiscipulus *Python scikit-learnRRapidMinerSalford Predictive ModelerSAS Enterprise MinerShogun toolboxSTATISTICA Data MinerWeka-----------Libraries and Tools-----------Accord.NETILNumericsMath.NET NumericsWintellect Power CollectionsQLNetNoda TimeR.NETALGLIBLIBSVMRef: http://en.wikipedia.org/wiki/Machine learning5/6/2015
WHAT ARE THE BEST APPLICABLELANGUAGES OR LIBRARIES?5/6/2015 #CC C .NETC AMPOPEN MPELPLWLSFORTRANVERILOGVHDLCUDAOPEN CLCILKCLOUREHTML5R
REGISTER GENETICPROGRAMMING Based loosely on biological models of evolution and eucaryotic* sexualreproductionSimulates the path a biological species goes through as it evolves:-Starts off simple-Adapts to hostile environment-Strong Parents give birth to strong children-Random mutations may helpWorks at the FAST CPU Register Level, not high level codeFast, Accurate, and Writes CodeDifferent from GA and Tree Based GP*Based on complex cells with membranesReference: 0How%20It%20Works.pdf5/6/2015
TSL GP LEARNING Supervised Learning. No supervisory Signal.Population is initializedTrading Strategies are initialized with random signalsTournament is run within population applied to the trading simulatorMutation causes random changes in winnersCrossover exchanges DNA between winnersReproduction is applied on remainderDemes enhance genetic diversityParsimony Pressure favor simpler solutionsIf n GWI occur then run restartsNew trading algorithms emerge and improve based on the errorAlgorithms learn to trade better as they trade in simulationAfter x runs or user termination, all runs stopFinally, code is exported, translated and ported to a Trading OMS/EMS5/6/2015
EVOLUTIONARY BASEDINDUCTION OF MACHINE CODETSL’s Patented GP is 60-200 times faster than other algorithms.Present: FastPast: SlowCPUCPUStep 2. EvaluationMachine Code InstructionsDirect Execution. Nointermediate steps.Step 1. Convert tomachine codeData StructureMachine Code InstructionsEVOLVES DIRECTLY ON TRADING SYSTEM us%20How%20It%20Works.pdf5/6/2015
LAIMGP REPRODUCTIVECROSSOVERHomologous and Non-Homologous crossoverTradingAlgorithmTrading System canvary its size duringevolutionReference: Frank D. Francone Licensiate Thesis (2009)5/6/2015
FUNCTION SETS: DNAMore Function Sets allow deeper and wider ranges of solutions to be exploredTSL’s GP is 60-200 times faster than other AlgorithmsTSL uses 34 Function Sets including ulus%20How%20It%20Works.pdf5/6/2015
MACHINE EVOLVED AND WRITTEN CORELOGIC OF YOUR TRADING SYSTEMTranslation Path:Machine Code - Core Logic C Code - C#, EasyLanguage and otherslong double f[8];long double tmp 0;int cflag 0;f[0] f[1] f[2] f[3] f[4] f[5] f[6] f[7] 0;L0:L1:L2:L3:L4:L5:L6:L7:L8:L9:L10:L11:f[0]- v[25];f[0] v[43];f[0] fabs(f[0]);f[0]- v[13];f[0]- v[49];f[0]- v[41];f[0]* f[0];f[1]- f[0];f[0] v[22];tmp f[1]; f[1] f[0]; f[0] tmp;cflag (f[0] f[2]);f[0]- v[39];if (! finite(f[0])) f[0] 0;return f[0];5/6/2015C#, EL, PL, BLOX ormany other platformsNote only 7 inputs are used here out of theInitial 56 fact Terminal Set available
RML’S DISCIPULUS, USED EXCLUSIVELY IN TSL,OUTPERFORMED EXISTING PUBLISHED RESULTSUXO Discrimination TestsDiscipulus
TSL MAIN COMPONENTSLEARNINGMACHINE5/6/2015TRADINGSIMULATOR
A TRADING SYSTEM MAPS DATA TOEQUITY CURVESTheoreticallyPerfectEquity CurveDATAProfitorLossMACHINEDESIGNED SYSTEMMAPPINGTimeThe resultant equity stream net profit np[n] is given by:𝑛𝑝[𝑛] 𝑛𝑡 1𝑡𝑝 𝑡The resultant net profit at t is given by: opp𝑛𝑝[𝑡] 𝑛𝑝 𝑡 1 𝑡𝑝[𝑡] opp5/6/2015
TSL INPUT PREPROCESSING10 Built In PP’s. Open Code-Fully customizable. 56 Inputs availableClassical and Non-Classical Patterns 1, 2 and more bar patterns Momemtum Patterns Countertrend Patterns Trend Patterns Gaps and variations Adaptive boolean patterns Adaptive numeric pattern relationships Support and Resistance, adaptations and variations Detrended pattern effects and variationsClassical and Non-Classical Indicators Normalized variables Transforms Standard Deviation and variations Averages and variations Volatility, Volatility Ratios and variations Adaptive Channels Regressions and variations Oscillators and variations Detrended prices, oscillators and variationsOther DNA: Intermarket data Fundamental data COT Machine readable news Social Media Exogeneous Data Order Book Bid/Ask & Size Order Book Movement Order Book Stats
SOCIAL MARKET DATATSL AND EOTPRO DEVELOPING TECHNOLOGIESWithout Social Media DataWith Social Media Datahttp://www.eotpro.com/5/6/2015
TSL SIMULATION ROUTINES 25 Trade Types including multi-systems40 Fitness Functions - External FF API DLL56 Inputs8 Outputs11 Risk, Size, Stops, Targets5 Preprocessors 5 IDEVORUN 5/6/2015
THE STUDY OFENTRY TYPESEach order type has many variations.Which one has the best EV for yourMarket under Study?Counter TrendBreakoutMarket Order(TSL has 25 Trade Types)5/6/2015
MARKETS HAVE DIFFERENTDESCRIPTIVE STATISTICSSo Why Design Symmetrical Systems?TSL will design systems within systemsPower Spectral DensityIndicator Serial CorrelationRandom TrendCME:E-MINI S&P5/6/2015CBOT:WHEAT
EVORUN ON SETUPS (TT) Trade Types/Trading Tactics are entry or order techniques.Example: Enter on Limit(FF) Fitness Functions are “Targets” that TSL attempts to design to.Example: NetProfit/Max Drawdown(PP) Preprocessors are Patterns, Indicators or other facts used asDNA in TSL. Example: Close Close[1](TTPD) Trades Per Day. Determines efficient intra day trading frequency.(BS) Bar Size. Optimum bar size needs to be stochastically determined.There are 25 TT’s, 40 FF’s, 10 PP’s many TTPD’s and BS allowing millionsof possible setups to be tested, each allowing millions of systems to begenerated for each setup.Clearly there is a need for simplicity and runs reduction5/6/2015
EVORUN WHAT IS THE BEST BAR SIZE?ES 15 Min?5/6/201530 Min?45 Min?60 Min?
WHAT IS EVORUN ?EVORUN is a TSL multi run iterator:1.2.3.4.5.5/6/2015Trade TypeFitness FunctionPreprocessorBar SizeMax Trades per Day
TSL MAIN COMPONENTSLEARNINGMACHINETRADINGSIMULATORFITNESS EVALUATOR5/6/2015
FITNESS CAN BE MULTI GOALMachine Design Allows Us to Adjust Critical System Metricsas Targeted Fitness FunctionNet ProfitDrawdownPercent AccuracyExample fitness:𝒏𝐀𝐌𝐑 (𝟏/𝒏)Profit FactorAverage Trade5/6/2015𝐏𝐎𝐒(𝐌𝐂𝐄 𝐢 𝐓𝐄 𝐢 )𝒊 𝟏
SYSTEMS AS OBJECTS:FITNESS AS EXPRESSIONSNetProfitNetProfittimetimeNotice drawdown improved even though DD was not part of fitness5/6/2015
FITNESS GAMESIncreasing Run TimeREWARDTORISKDynamic Fitness orDNA injectionCode beginsto usedifferentDominantDNA setsCode usesDominantDNA set 1Good OOSGood TRNApproach #1Emerges.Comparableto basicstrategiesBetter OOSBetter TRNApproach #2Emerges.Comparable toIntermediatestrategiesDESIGN TIME GENERATIONSHow, what, where, when, why it learns5/6/2015AlternativeDNA setsemerge intoselectionBest OOSBest TRNAlternativeApproaches #3Emerges.Comparableto advancedstrategies
TSL MAIN COMPONENTS9 Languages, 1 million lines of code, 2 companies, 10 years in developmentEXCHANGE DATAEXOGENOUS chine Codeto C to C#, EL,etc.FITNESS EVALUATORRT PERFORMANCEEXCHANGEDATAORDERS5/6/2015MACHINE DESIGNEDTRADING ALGORITHMSOMSEMS
5/6/2015
A DISTRIBUTION OF UNIQUESYSTEMS FROM TSLSystems are uniqueand novel. Evolvesdifferent systems foreach user even withsame setup due toStochastic nature ofprocess.5/6/2015
A WORD ON SLIPPAGE ANDCOMMISSION IN TSL-GPAverage Trade is often the limiting issue as trading frequency increasesUsing S&C at the start of the run prevents potential good material from evolvinginto very acceptable end solutions and structures, so apply S&C at the end of the run.Acceptablewith S&CAcceptablewith S&CEvolutionCannot overcomeInitial S&C debitS&C USED AT STARTOF andFITNESStime5/6/2015S&C USED AT ENDOF RUNSWithout S&Cevolution progressesmore efficiently but finalsolution is acceptablewith S&Ctime
ROBUSTNESS(Over Fit Avoidance) Forward and Back OOS Testing (Walk either)Run Path Logs (Path intelligence)Unbiased Terminal Set (Directionless inputs)Multi-Run, Randomized Criteria (Global optimum)Zero Point Origin (No predefined initial point)Parsimony Pressure (Occam’s razor)Stat Tests-Distribution is exported (Reject Null)TTPR (Degrees of Freedom)Data duration and choice (More is better)Post Design/Post OOS tests (Second Blind)Sequestered Data Testing (Extreme testing)
A FEW OMS/EMS THAT CAN HOSTTSL TRADING STRATEGIES TradeStation (EL)MultiCharts (PL)Deltix (C#)QuantHouse eMarketData(C )NYSE API(C )Mantara (C#)WaveRules (C )AB2000 (C )Trading Blox(Blox)Ninja(C#)WealthLab (WLS)Others via native languages or TSL DLL5/6/2015
TRADING STRATEGY DESIGNREDUCED TO 3 SIMPLE STEPSNo Programming Required1. Preprocess2. Evolve3. Translate5/6/2015
PROBLEMSEVERYWHERE The world is getting smaller and interconnected Inter-market Correlations are increasing Co-holding risk is largely hidden, expensive to hedge anddifficult to diversify against The markets keep changing dynamically while systems are notadaptive enough, difficult and expensive to create Returns are elusive and require deeper mining The worst risks are potentially unknown and unknowable Money Managers are typically slow to adapt
TSL MAJOR PROJECTS TSL - Current Commercial Platform ProductLimited External Strategic ConsultingInternal R&D supporting:Quant Systems Lab (TSL Next Gen)5/6/2015
WHAT IS THE BOTTOM LINE?There is no way a Human candesign as many unique and novelTrading Strategies as TSL and then have the code writtenfor you5/6/2015
CONCLUSIONMACHINE LEARNING IN TRADING WILLCONTINUE Check out the Kindle Book: Best Trading Strategies andour section on Machine Designed Trading StrategiesJoin our Silicon Valley Machine Learning for Trading Strategies MeetUp Group5/6/2015
TSL RECENT ADDITIONS New Fitness Functions :Net Profit/Average Max DDNet Profit/Average Trade DurationExternal API User Defined e.g.:If Fitness Calls 100000 Fitness Net Profit/ Avg Max DDelse Fitness NetProfit Multi Asset Robo Advisor:Long Only PortfolioShort Only Portfoliowith Constant Dollars GUI Enhancements:Quick Save/Quick Load SettingsSave/Load Any Settings5/6/2015
TSL RECENT ADDITIONS EVORUN:Performs run iteration on:BarSizeTradeTypesPreProcessorFitness FunctionMax Trades Per Day SOFT Fitness Targets:Average TradeMAX Drawdown QUANT SYSTEMS LAB5/6/2015
SOLUTION SPACEWhat do you do for each of these cases? 5/6/2015Poor TRN, Poor OOSGood TRN, Poor OOSPoor TRN, Good OOSGood TRN, Good OOS
WHAT IS TSL’S MACHINEDOING NOW? 5/6/2015TSL is Learning to trade better as it is trading in the SimulatorSystems are being tested OOS DURING EvolutionParameter reduction is automaticStrategies are being simplified due to Parsimony PressureEquations are being written and manipulatedStrategies are improvingMachine code blocks are being manipulated in FPU’sFinally code is up translated from register machine code.
Predictive Modeling (ML) Self-Evolving Strategy Structure (ML-TSL) . IBM SPSS Modeler KNIME KXEN Modeler LIONsolver MATLAB Mathematica mlpy MLPACK library MCMLL OpenCV dlib Oracle Data Mining Orange Discipulus * Python scikit-learn R RapidMiner Salford Predictive Modeler SAS Enterprise Miner Shogun toolbox STATISTICA Data Miner Weka .
