Equity forecast using Machine Learning - MUTIS
Post on 16-Apr-2017
PowerPoint PresentationForecasting equity using Machine LearningNikola MiloeviGoal Copyright 2016 MUTIS. All rights reserved 2016.Predict long term equity price movementOne year periodClassify which equities will grow by 10%Past data are knownFocus on technical analysis2Traditional approach Copyright 2016 MUTIS. All rights reserved 2016.Graham criteriaStock Selection for the Defensive Investor:1. Not less than $100 million of annual sales.[Note: This works out to $500 million today based on the difference in CPI/Inflation from 1971]2-A. Current assets should be at least twice current liabilities.2-B. Long-term debt should not exceed the net current assets.3. Some earnings for the common stock in each of the past 10 years.4. Uninterrupted [dividend] payments for at least the past 20 years.5. A minimum increase of at least one-third in per-share earnings in the past 10 years.6. Current price should not be more than 15 times average earnings.7. Current price should not be more than 1-12 times the book value.Graham number = sqrt(22.5*EPS*BV)3Other approaches Copyright 2016 MUTIS. All rights reserved 2016.Models inspired by GrahamsFollowing news and trends4Problems with Graham model Copyright 2016 MUTIS. All rights reserved 2016.It was developed in 1940sIt is hard to find a stock that satisfies criteriaToo strictToo defensive5Help from technology Copyright 2016 MUTIS. All rights reserved 2016.In past decade were developed approaches based on technologyAlgorithms based on statistics, heuristics, probability and machine learningThey mainly focused in the past on short term trading6Machine learning intro Copyright 2016 MUTIS. All rights reserved 2016.Field of study that gives computers the ability to learn without being explicitly programmed7Experiment (1) Copyright 2016 MUTIS. All rights reserved 2016.Use machine learning on past 2-3 year dataData obtained using Bloomberg terminalData include 28 indicators Book value, Market capitalization, Change of stock Net price over the one month period, Percentage change of Net price over the one month period, Dividend yield, Earnings per share, Earnings per share growth, Sales revenue turnover, Net revenue, Net revenue growth, Sales growth, Price to earnings ratio, Price to earnings ratio -five years average, Price to book ratio, Price to sales ratio, Dividend per share, Current ratio, Quick ratio, Total debt to equity, margins, asset turnover8Experiment (2) Copyright 2016 MUTIS. All rights reserved 2016.Selected 1739 stocks from different indexes (S&P 1000, FTSE 100 and S&P Europe 350)Calculated which ones price grew more than 10%Used different Machine learning algorithms and 10 fold cross validation for evaluationUsed Python for scripting and Weka toolkit for machine learning9Results (1) Copyright 2016 MUTIS. All rights reserved 2016.Trial with all financial indicators as a features10Results (2) Copyright 2016 MUTIS. All rights reserved 2016.We performed feature selection among the indicatorsExperiment with only 11 indicators1111 indicators that were good Copyright 2016 MUTIS. All rights reserved 2016.The performance turned out not to be significantly different, but it showed that only 11 indicators are enough12Best performer Copyright 2016 MUTIS. All rights reserved 2016.13Decision trees (1) Copyright 2016 MUTIS. All rights reserved 2016.Tries to understand the data and build a decision tree based on data14Decision trees (2) Copyright 2016 MUTIS. All rights reserved 2016.OutlookSunnyOvercastRain15Decision trees (3)OutlookSunnyOvercastRainHumidityHighNormalDont playPlayWindWeakStrongPlayDont playPlay16Random forests Copyright 2016 MUTIS. All rights reserved 2016.Algorithm that creates a forest of decision treesDesigned to improve the stability and accuracy of machine learning algorithmsReduces variance and helps to avoid overfittingUses technique called bagging17Bagging Copyright 2016 MUTIS. All rights reserved 2016.From a set of elements, creates n sets of elements (in our case randomly)Builds n models using subsets for each modelIn order to get final class uses voting strategyClass with majority of votes wins18Example Copyright 2016 MUTIS. All rights reserved 2016.19Reference Copyright 2016 MUTIS. All rights reserved 2016.Milosevic, Nikola. "Equity forecast: Predicting long term stock price movement using machine learning."arXiv preprint arXiv:1603.00751(2016).https://arxiv.org/ftp/arxiv/papers/1603/1603.00751.pdf 20Thank you and questions Copyright 2016 MUTIS. All rights reserved email@example.com 21
View more >
Drugs store sales forecast using Machine 229 Final Project Report! Drugs store sales forecast using Machine Learning Hongyu Xiong (hxiong2), Xi Wu (wuxi), Jingying Yue (jingying) 1 Introduction
Spoken Language Identification using Machine LANGUAGE IDENTIFICATION USING MACHINE LEARNING ... May 2012 . Department ... Spoken Language Identification using Machine Learning
FINANCIAL TIME SERIES FORECASTING A MACHINE LEARNING ... TIME SERIES FORECASTING A MACHINE LEARNING APPROACH ... to forecast financial time series.