Successful traders know how important it is to have a well-defined strategy. It is also very important to have confidence in this strategy. To acquire this confidence, traders perform what is called backtesting. Backtesting a strategy is the process of simulating this strategy over the past given historical data. This simulation will tell you whether your strategy is profitable and how well it has performed in the past. This is done using computer algorithms and trading software programs (like QuantShare). Because the whole process is depending on the historical data you have, there are many common pitfalls that you should be aware of and that you should avoid. First, the historical data must be adjusted for all corporate actions, which includes splits, dividends, bonuses... Data providers generally deliver adjusted historical quotes, so this is not really an issue. Other pitfalls include Look-ahead bias, Data-snooping bias, Survivorship bias... The Survivorship bias is one of the most important, so let's dig deeper into this bias. First, let us explain with an example what the Survivorship bias is. Your historical database contains quotes for about 8000 ticker symbols. All these companies still exist. When you perform a backtest that spans from 2000 to present, your simulation will buy and sell stocks among these 8000 symbols. However, during the 2000-present period, many companies have disappeared for different reasons. The simulation will not buy and sell these stocks and this is what causes the Survivorship bias. Consequence: An import consequence of this bias is that your simulation performance may be inflated (this is especially the cause for "value" strategies). This is because; these disappeared companies are not included in the simulation while a lot of them have performed poorly, especially companies that have gone bankruptcy. In other words, the Survivorship bias will cause your simulation to include only companies that were successful enough to survive until the end of the simulation period. Going bankruptcy is not the only reason why a company disappears; several other reasons exist, this includes delisting, mergers, acquisitions... Solution: The solution is to have a survivorship bias-free database. That is, a database that includes also the disappeared companies, as well as their historical prices. This type of database is also called "point-in-time" database. Many data providers don't have such data. For those who do, they generally sell it separately. In QuantShare, we have a downloader that gets this data for free. Here is the downloader link: Historical Quotes for Delisted US Stocks. You should also download a list of disappeared companies' tickers at the following link: US delisted symbols.
|
|
|
|
|
