Machine Learning for Trading Financial Investing Technical - PowerPoint PPT Presentation

Efficient Markets Hypothesis (does not support our assumption) Assumption: We can gain advantage in the market from exploiting different sources of ‘ information ’ Machine Learning for Trading Financial Investing Technical Analysis: • Historical Price (movements – are not random). The Fundamental Law • Volume of active portfolio management Fundamental Analysis: • Features of the intrinsic value of a stock, e.g., earning. Intuition to Earning Money on the Stoc General Investment Intuition Market News (Information) of a Company Possible Scenario 1. Eric learns good news 2. Eric tells his grandmother Berta 3. Berta buys stock in advance 4. Announcement to public Good news Bad news Berta has an advantage • Stock price goes • Stock price goes up! down! • Good investment • Poor investment

Efficient Markets Hypothesis • All relevant information flows instantly - or super quickly – no one can take advantage of slow flowing information to gain an advantage. • Any information is available instantly in a [perfectly] efficient market. • Reflect: Both Fundamental & Technical Analysis are based on information so they are Efficient Market Hypothesis useless in a [perfectly] efficient market. • Instant Information Flow – Eric, Grandmother, and Public Learn about the news at the same time. Information : Efficient Market Hypothesis From public to less public • Jules Regnault, 1863 • Price volume (most public) • Eugene Fama, 1960s (formalized the idea) • Fundamental (intrinsic value) • Exogenous – other or related information affecting price that is • Stocks trade at their fair value not intrinsic information – impossible for investors to either purchase • Example price of oil may affect a company making undervalued stocks or sell stocks for inflated cars. prices. • Insider Information. https://en.wikipedia.org/wiki/Efficient-market_hypothesis

Efficient Market Hypothesis Efficient Market Hypothesis Forms Forms • Weak • Weak – Future price cannot be predicted by analyzing – Future price cannot be predicted by analyzing historical prices historical prices à Technical Analysis cannot work à Technical Analysis cannot work • Semi-Strong • Semi-Strong – Prices adjust rapidly to new public information – Prices adjust rapidly to new public information à Fundamental Analysis cannot work à Fundamental Analysis cannot work • Strong: • Strong: – Prices reflect all information, public an private. – Prices reflect all information, public an private. à No analysis relying on ANY information (including à No analysis relying on ANY information (including insider information) cannot work insider information) cannot work Efficient Market Hypothesis Efficient Market Hypothesis Forms Forms • Weak • Weak – Future price cannot be predicted by analyzing – Future price cannot be predicted by analyzing historical prices historical prices à Technical Analysis cannot work à Technical Analysis cannot work • Semi-Strong • Semi-Strong – Prices adjust rapidly to new public information – Prices adjust rapidly to new public information à Fundamental Analysis cannot work à Fundamental Analysis cannot work • Strong: • Strong: – Prices reflect all information, public an private. – Prices reflect all information, public an private. à No analysis relying on ANY information (including à No analysis relying on ANY information (including insider information) cannot work insider information) cannot work

EMH Prohibits EMH Prohibits Semi Semi Weak Weak Strong Strong Strong Strong Technical Technical þ ����� þ ����� þ � ����� ����� � � � Fundamental ����� þ ����� þ Fundamental ����� ����� Insider ����� ����� þ Insider ����� ����� Case Study (Estimating Value of a Stock) • Look at Price Earning • Stocks moves in direction of Ratio their earning over time – 33% of investment • PE ratio compares price of stock managers considers this to its recent earning. ratio before buying stock. • Making money then stock price – 15-20 PE Ratio (AVG will eventually go up and (low PE 15.54), typical on average ratio, higher returns, low expectations, easier to – Question: What type of information is the PE overcome) Ratio? (Technical, • Losing money it’s stock price will Fundamental or Insider?) go down (high PE ratio, lower • Lower PE Ratio is better returns) – Intuition: Low PE Ratio means lower expectation https://jbmarwood.com/historical-pe-ratios/

Fundamental Law of Active Portfolio Management • Low PE Ratio is better • Warrant Buffet: • A clear correlation – Wide diversification is only necessary when the – Lower PE ratio is equal to investors do not know what they are doing. better investment returns – Holds through each twenty • Skill & Breath year period. • Challenges: – Skill – Selecting the right stocks. – Stock market will adjust – Breath – number of investment opportunities – Historical PE ratio not easy • Grinold’s Fundamental Law: to come by. – Performance = Skill * √ Breath Casino Casino Coin flip: Which is better? Coin flip: Which is better? • 1 bias coin • 1 bias coin (Probability of Winning) x (Amt Won per Bet) – .51 heads à Win. – .51 heads à Win. – (Probability of Losing) x (Amt Lost per Bet) – .49 tails à Lose. – .49 tails à Lose. Bias – is our edge, our skill. • 1,000 tokens that you can bet: • 1,000 tokens that you can bet: • Which is better ? • Which is better ? – 1,000 tokens that you can bet: – 1,000 tokens that you can bet: • Bet 1 : 1 bet of 1,000 tokens • Bet 1 : 1 bet of 1,000 tokens • Bet 2: 1000 separate bets, one at a time. • Bet 2: 1000 separate bets, one at a time. • Bet 3: Both are equivalent? • Bet 3: Both are equivalent?

Expected Return Risk: Take 1 – Lose it all. • Single Bet: • Lose it All – 0.51 * $1,000 + 0.49 * -$1000 – Bias Coin à .49 Losing. – 510 – 490 = $20.00 Profit. • Single Bet • Multi Bet: – .49. – (0.51*1.00) – (0.49*-1) = .51-.49 = .02 c. • Multi Bet – Make the bet 1,000 times: .02*1,000 = $20.00 – .49 * .49 * .49 … – [.49] 1000 = really small chance you lose it all. Expected Return is the SAME Risk: Take 2 – Standard Deviation. Coin flip: Reward/Risk • Allocating bets differently across tables. • Combine – 1 Extreme bet it all at one table – Expected Return & Risk • AT One table: • Similar to the Sharpe Ratio – Win 1,000, or Lose 1,000 • AT Other tables (we did not bet on these) • Reward/Risk à Expected Return/ StDev() – Outcome is 0. • Stdev(1000,0,0,0, … ) = 31.62 – 1 Extreme evenly distribute the bets 1 bet at each table. • 1 win, -1 lose • Stdev(-1,1,-1,1, … ) = 1. • Single Bet: • à standard deviation is 1. – 20/31.62 = 0.63 – Then there are the in-betweens. • Multi Bet: • Summary: – Risk / Standard Deviation is much larger if we do 1 single Bet. – 20 / 1 = 20.

Towards a Model. • Single Bet: • Same relationship in active portfolio – 20/31.62 = 0.63 (SR single ) management. • Multi Bet: • SR multi = SR single * SQRT( 1,000) – 20 / 1 = 20. • Performance = Skill * SQRT(Breath) • Combine Single bet with Multi Bet • Lessons: • SR multi = SR single * SQRT( 1,000 ) – Higher Alpha (skill) à Higher SR • Performance improves with Breath. – 1 single bet no improvement – More Opportunities à Higher SR – 1000 spread maximum improvement. – SR grows with the SQRT(Breath). Fundamental Law • Renaissance Technology • IR = IC * SQRT (BR) – 100K/day • IR – information Ratio • Warren Buffet • IC – Information Coefficient – 120 Stocks • BR – number of trading opporturnities • Information Ratio/Reward. – Alpha encapsulate skill – Recall : R p (t) = Beta R m (t) + alpha – IR = mean(alpha)/stdev(alpha)

Efficient Frontier. • Constraint in optimizer is a particular expected return (recall in our optimizer the constrain was that allocations summed to 1) • Markovitz Bullet • Maximum Sharpe Ratio