What are we talking about when we talk about outliers? We’re talking about extreme or unexpected values in market data which may or may not represent real events. For example, if we’re tracking a solar energy company stock price, we might see unexpected movements in the price on a day when a solar eclipse occurs. Outliers can show up for a million different reasons but some reasons are more common than others. In this video, we’re going to talk about a few of the most common types of outliers. Outliers can always appear due to human error, either during manual data entry or in the form of computer bugs. These are known as fat finger errors. To cite a few examples that had newsworthy consequences, in 2001 a Japanese traitor cost his employer UBS Warburg, 71 million pounds when he sold 610,000 shares at 60 yen instead of six shares at 610,000 yen as he had intended. In 2015, a junior member of Deutsche Bank’s foreign exchange sales team processed a trade using a gross figure rather than a net figure causing a payment to a U.S. hedge fund of six billion dollars, orders of magnitude too high. Sometimes data can be missing completely, entered as zeros or duplicated from previous states. This can happen due to lax quality control of market data vendors but sometimes exchanges themselves can be missing data. Data maybe missing because the stock was suspended from trading for a day or more often, for part of a day. Trading on a stock can be stopped for regulatory reasons, to reduce volatile trading prior to a companies news announcement or when there’s uncertainty about whether the stock continues to meet listing standards. It can also be stopped for non-regulatory reasons. For example, if there’s a significant imbalance and pending buy and sell orders. In these cases, one way to check if trading actually occurred is to check if the trade volume was zero. Trading for an entire exchange can also be stopped. There are mechanisms called trading curves or circuit breakers that halt trading if an index drops by a certain percent. The idea is to prevent stock market crashes, events when the prices of stocks across a significant cross section of the market drops suddenly and significantly. You might also see extreme price movements that are due to real events such as earnings, mergers or other announcements. Announcements represent information that the market previously did not know. So these can surprise the general market if the results are much better or worse than the market expected, so to speak. When the market adjusts to this information, the stock price fluctuates as the new information starts to be reflected in the trading price. For example, the athletic apparel company, Puma, made an unexpected earnings announcement on October 18th, 2017 in which it reported stronger than expected third-quarter earnings. That day, its stock closed four percent higher than the previous day’s close and the subsequent days saw fluctuations as the price rose still further. Outliers can occur due to the intended or unintended actions of computer trading programs. For example, in 2010 the market experienced a flash crash which lasted only 36 minutes and was later linked to trading algorithms. While extreme price fluctuations occurred during this short period, they were due to real events. Sometimes quants must work with unadjusted or nominal price data because adjusted price data are unavailable through their data source, when working with real time data directly from a stock exchange for example. If you are working with unadjusted price data, you will see discontinuities in prices on dates when companies issue dividends, split their stock or took other corporate actions. Companies may issue dividends on a regular schedule or may decide to issue one-off so-called special dividends. In this example, you can see that in 2017, Costco issued several regular dividends as well as a large seven dollar per share special dividend. These discontinuities can look like outliers if returns are calculated on these data. A stock trades ex-dividend on or after the ex-dividend date, so on this date, non-adjusted prices decrease. This represents the fact that the net company value has decreased because cash has been transferred from the company to shareholders. However, after prices are adjusted, change demand for company stock may also cause prices to change. Price changes due to change demand are real market effects. To deal with discontinuities in unadjusted prices, whoever is cleaning the data must go back, look at corporate events and cumulatively adjust for each dividend distribution to generate adjusted prices.