Data Historical Machine Learning for Trading Price Volume - PowerPoint PPT Presentation

Data • Historical Machine Learning for Trading • Price • Volume Financial Investing Dealing with Data How Data is aggregated How Data is aggregated 10020 10000 • Many trades • Many trades 10000 9998 9997 9996 9995 9991 9989 9982 9980 9980 9977 9979 • Different Exchanges • Different Exchanges 9966 9965 9961 9960 9960 9959 9958 9959 9956 • Tick • Tick 9947 Price 9944 9940 Price 9939 9935 9940 9937 – Finest resolution of 9936 – Finest resolution of 9929 9929 9920 9919 9916 data (minimum size) 9921 data (minimum size) 9920 9907 9917 9917 9900 • Price (1c) • Price (1cent) 9900 9880 Volume: 200 100 300 100 200 100 300 100 500 100 Volume: 200 100 300 100 200 100 300 100 500 100 • Upward, downward • Upward, downward 9860 movement (change in 9880 movement (change in 1/23/18 2/23/18 3/23/18 4/23/18 5/23/18 6/23/18 1/23/18 2/23/18 3/23/18 4/23/18 5/23/18 6/23/18 price) price) Time – Cell Match – Cell Match • Buy/Sell Match (not guaranteed at specific • Buy/Sell Match (not guaranteed at specific times) times) • Buy and Sell Match • Buy and Sell Match Different Exchanges. Different Exchanges. • •

Price anomaly • Lots of Data • Example 1: • Consolidate Data – Blue Line: – Time Epoch – What are the • Minute by Minute price drop here? • Hour by Hour • Daily – Open, Close – High, Low – Volume – Combine different exchanges Price anomaly Price anomaly Which makes most sense? • Example 2: – And here, what are q CEO Quit these price drops? q Dividend Cut q Stock Split

Price anomaly Why do Stock Splits? • Price becomes to high: less liquid, less volume. • Buy in groups of 100 (typical) q CEO Quit – 50,000 for $500/stocks. • Exceptions/Why make an Exception? q Dividend Cut – Berkshire Hathaway BRK.A.(doesn’t split it q Stock Split shares). • Warren Buffet’s holding company. • >$110,000 (Volume : 450 shares per day) • Keeps the price high to deter short time traders • Next day 4 shares of creating excess volume. 75, 4x75=$300. – Seabord SEB, $2,660 • Next day 2 shares of – NVR $703 125 – 2x125=$250. – GOOG -- $618 no splits. Started at $100 in 2004 2017 prices https://www.forbes.com/sites/investor/2011/07/25/bershire-seaboard-google-priceline/#78528b8d64ab Why Split? • How to deal with • Less Liquid, less volume stock split data • Options on stocks are usually traded with • Not short stocks when regard to 100 shares. company value has • Finely Tuned portfolio (harder with high not change: priced stocks). – Adjusted Close. this situation. This green line represents the actual price – Current Day: • Adjusted close = close • Back in history need to adjust it.

Dividends • How to deal with • Dividends, stock split data – Annually – Quarterly • Not short stocks when • Example company value has – $100.00 not change: – $1.00 dividend – Adjusted Close. announced (1%) this situation. This green line represents the actual price – Stock value? – Current Day: Here we have: • Assume consensus • Adjusted close = close $1 + $100 share of stock vale is 1) What is the value of the • Back in history need to $100.00 (may not stock the day before be the price). adjust it. dividend payout? 2) What is the value on they day of payout? • Day before • Answer: and day off? – $101 day before – $100 day after.

Adjusting for Dividends Survivor Bias • SP 500 • Similar to stock – Membership Change. • Simulate todays list in the past using splits. historical data • Red line is – Current list survived in the past. Adjusted close. • Survivor Bias- – Inflation of funds that remain when poor performers are not part of the equation. • Need survivor bias free data (costs money). – To more accurately measure strategies working on historical data. – Need to know historical lists of companies. https://www.elitetrader.com/et/threads/can-anyone-recommend-a-source-of-historical-data-without-survivor-bias.212720/