Adjusted data
- Thread starter azouz
- Start date
Murray Ruggiero
Sponsor
Doing backtests using split adjusted data has issues because you can only really test on one stock at a time. This is because you destroy the price relationship of the stocks in a portfolio. We have done a study and Yahoo data has problems before 1992. After 1992 ,CSI started supplying data to them from what I understand. Yahoo only supplies split adjusted data.
TradersStudio uses unadjusted, split adjusted and now dividend adjust and create a special stock format which our backtester uses and retains the original price information and also all the correct trades, handing the issues caused by splits in a different way. Please read this tutorial from our site. This discusses the old algorithm without dividend adjustment. The link to the free tutorial is below.
http://www.tradersstudio.com/Default.aspx?tabid=126
If you register on the site which is free you will get a 20+ page PDF which will include rules to the system.
TradersStudio uses unadjusted, split adjusted and now dividend adjust and create a special stock format which our backtester uses and retains the original price information and also all the correct trades, handing the issues caused by splits in a different way. Please read this tutorial from our site. This discusses the old algorithm without dividend adjustment. The link to the free tutorial is below.
http://www.tradersstudio.com/Default.aspx?tabid=126
If you register on the site which is free you will get a 20+ page PDF which will include rules to the system.
Whether or not to use split-adjusted data depends on assumptions made in back tests. Most significantly, split-adjusted data may be used if we assume a fixed-dollar position in each trade (e.g., $5000) rather than a fixed number of shares. In such case, relevant relationships remain valid and testing across multiple stocks is possible.
how do you handle this with unadjusted data :
you do a backtest and the results show you this :
You enter lets say AA stock at 20$ and sold it after one month at 10$, a -50% performance.
But in this month AA was splited and with adjusted results, the real performance will be 0%
this can cause a lot of error in the backtesting
you do a backtest and the results show you this :
You enter lets say AA stock at 20$ and sold it after one month at 10$, a -50% performance.
But in this month AA was splited and with adjusted results, the real performance will be 0%
this can cause a lot of error in the backtesting
Murray Ruggiero
Sponsor
Not really, because you if you assume $5000.00 then that not correct because you might not have been able to buy $5000.00 because it would have give you fractional shares. Another issue is that the rounding becomes a problem. When Microsoft split adjusted is at .1100 , with a real price of $34.00, then errors of .0001 results in about a .034 error in real life. If you compound this over 1000's of trades and 1000's of shares it is a real problem.
In addition you can't apply even basic money management strategies, for example sizing your position based on trade risk on a technical level, for example a stop at the 50 day moving average. This is because you can no longer assume you are buying the same amount of each stock.
Another example would be you could not use a price filter for example don't trade stocks under $5.00.
I don't know of any backtesting , portfolio based platform which handles the issues of portfolio based stock analysis and money management on them correctly. The only software I know of which handles this analysis correctly is TradersStudio.
Fixed-dollar position testing is not without its complications. Fractional shares is a factor to consider. The related distortions become more pronounced as stock prices increase and trade amounts decrease. For example, trade size = $500 and stock price = $45 will stumble over this problem more so than trade size = $10,000 and stock price = $4.50.
I struggle with the other examples cited as problems related to fixed-dollar trades ...
Rounding errors may be more a theoretical concern than a practical one. For the problem to become significant, multiple splits would have to affect many of the stocks in back tests. Also, it seems a trader would also have to be evaluating very low positive expectancy systems to be concerned about rounding errors.
Fixed-dollar trades does not preclude use of all money management techniques. We could still define our maximum risk per trade, employ stop-loss strategies, and diversify our positions. On the other hand, we would not be able to adjust sizes based on volatility or equity changes.
I'm not sure why we couldn't use price filters (e.g., < $5) when using fixed-dollar trades. In fact, I do it without problem.
I may not clearly understand the process TradersStudio uses for handling split-adjusted data. Figuratively speaking, it appears to "unadjust" split-adjusted data, returning it to pre-split values. If that is correct, it would be understating the impact of transactions fees as a percentage of price. This is probably not significant, though, for reasons discussed previously for multiple splits.
Also, if I understand correctly, the "unadjustment" process requires open interest data. Not all pricing services -- or stocks -- have this information.
I struggle with the other examples cited as problems related to fixed-dollar trades ...
Rounding errors may be more a theoretical concern than a practical one. For the problem to become significant, multiple splits would have to affect many of the stocks in back tests. Also, it seems a trader would also have to be evaluating very low positive expectancy systems to be concerned about rounding errors.
Fixed-dollar trades does not preclude use of all money management techniques. We could still define our maximum risk per trade, employ stop-loss strategies, and diversify our positions. On the other hand, we would not be able to adjust sizes based on volatility or equity changes.
I'm not sure why we couldn't use price filters (e.g., < $5) when using fixed-dollar trades. In fact, I do it without problem.
I may not clearly understand the process TradersStudio uses for handling split-adjusted data. Figuratively speaking, it appears to "unadjust" split-adjusted data, returning it to pre-split values. If that is correct, it would be understating the impact of transactions fees as a percentage of price. This is probably not significant, though, for reasons discussed previously for multiple splits.
Also, if I understand correctly, the "unadjustment" process requires open interest data. Not all pricing services -- or stocks -- have this information.
Murray Ruggiero
Sponsor
<b> Answer to QQQShort </b>
You are correct that split adjusted data will give you the correct percent returns if you assume that you always buy $10,000 worth of each stock and forget about the issues with fractional shares. The problem is that this might require Excel to do this analysis on a large portfolio depending on your software. This means that you cannot optimize your portfolio on a large basket of stocks without running into issues of requiring custom programming.
You say in your write up you donât understand why you canât use a price filter for fixed dollars trades. My question is if you are using split-adjusted data where does the unadjusted price for filtering come from. TradersStudio stock contract use adjust, unadjusted and now optionally dividend adjust data to create a new price series. We use the open interest field to store additional information.
TradersStudio Stock contracts have data streams called TSOpen, TSHigh, TSLow, and TSClose in addition to the standard open, high, low, close. This allows you to access both a back adjust prices in the normal open, high, low, close fields and a raw price using these other fields. The reports use the raw prices and calculate the correct percent returns and dollar returns, even if you want to test trading 100 share of each stock. This might be doable without TradersStudio but would require a lot of custom programming, when TradersStudio does it automatically. Having real dollar profits and percent profits in the same analysis is very useful in evaluating your trading strategy. .
Letâs show a practical example. Letâs assume you want to know if you started with $10,000 and started trading on Jan 2,2000, how much money would you have on April 1, 2006. You want to test your system and money management strategy and try different what if scenarios. Since you have so little money you are screening stocks by only taking a maximum of 5 positions at a time and look to filter these positions using a relative volatility or strength filter ranking. Here another question, what about margin, if you margin yourself 150% how does that affects your risk-adjusted returns? In addition do you get a margin call, which in real life would have created a problem? What about if you wanted to lower your risk of going bankrupt and wanted to size your positions based on your account balance if all your protective stops are hit, so called core equity?. The list of the questions you can answer goes on and on and only TradersStudio can give you these answers in easy to use program.
You are correct that split adjusted data will give you the correct percent returns if you assume that you always buy $10,000 worth of each stock and forget about the issues with fractional shares. The problem is that this might require Excel to do this analysis on a large portfolio depending on your software. This means that you cannot optimize your portfolio on a large basket of stocks without running into issues of requiring custom programming.
You say in your write up you donât understand why you canât use a price filter for fixed dollars trades. My question is if you are using split-adjusted data where does the unadjusted price for filtering come from. TradersStudio stock contract use adjust, unadjusted and now optionally dividend adjust data to create a new price series. We use the open interest field to store additional information.
TradersStudio Stock contracts have data streams called TSOpen, TSHigh, TSLow, and TSClose in addition to the standard open, high, low, close. This allows you to access both a back adjust prices in the normal open, high, low, close fields and a raw price using these other fields. The reports use the raw prices and calculate the correct percent returns and dollar returns, even if you want to test trading 100 share of each stock. This might be doable without TradersStudio but would require a lot of custom programming, when TradersStudio does it automatically. Having real dollar profits and percent profits in the same analysis is very useful in evaluating your trading strategy. .
Letâs show a practical example. Letâs assume you want to know if you started with $10,000 and started trading on Jan 2,2000, how much money would you have on April 1, 2006. You want to test your system and money management strategy and try different what if scenarios. Since you have so little money you are screening stocks by only taking a maximum of 5 positions at a time and look to filter these positions using a relative volatility or strength filter ranking. Here another question, what about margin, if you margin yourself 150% how does that affects your risk-adjusted returns? In addition do you get a margin call, which in real life would have created a problem? What about if you wanted to lower your risk of going bankrupt and wanted to size your positions based on your account balance if all your protective stops are hit, so called core equity?. The list of the questions you can answer goes on and on and only TradersStudio can give you these answers in easy to use program.