PRELIMINARY MANUAL FOR THE RISK MANAGEMENT SPREADSHEET Version 0.5

Transcription

1 PRELIMINARY MANUAL FOR THE RISK MANAGEMENT SPREADSHEET Version 0.5 This is an early version of this program, many features of which have not been fully tested. It is free software released under the terms of the GNU General Public License, a copy of which is supplied with the spreadsheet. Users are warned that there is absolutely no guarantee of any kind associated with this software, even an implied guarantee of suitability for any purpose. See the GNU GPL for more details. This program was written as a tool for illustrating some important aspects of using stops and price-based buy signals as a way of protecting against losses, not as a generalpurpose tool suitable for investing. For that use, it will require more testing and the addition of features that reduce the possibility of user error. You should save a copy of the program is it is distributed before using it. It is not protected to commercial standards. In addition to this manual, the program is being posted with example data and results, as well as further Excel analysis on the Scratch worksheet. This illustrates some of the additional analysis that is necessary before the results from the optimization and backtesting functions can be integrated into an investing program. It is recommended that you look over that data before running your own analyses. The program was written with Microsoft Excel 2000 Visual Basic for Applications, and should be compatible with that version of Excel or later. It has been briefly tested with Excel 2003 and Excel 2007, and seems to work with both. Similar programs have worked with Mac versions of Excel, but this one has not been tested with a Macintosh. Please let me know via the address below if you have any comments or problems with the program. Bob Smithson The author can be contacted at: smithson@anava.com

2 General Comments on Back Testing and Optimization Past performance does not guarantee future returns. This statement is heard so often in the investing community that it known by virtually every investor. Yet it needs to be emphasized especially strongly in connection with computerized back testing and optimization. Any program that uses past data to test the performance of any investment strategy is particularly seductive, because the performance obtained by optimized mechanical strategies is extremely good over the period for which it is optimized. That is solely because the computer has perfect hindsight. It knows what the actual performance of all the tested investments was over the previous period! Unfortunately this kind of data is not available for the future. This exaggerated high performance tendency can cause disastrously poor results if it is not recognized and corrected for. Some people believe that the disadvantages of back testing can be eliminating by forward testing. Forward testing involves trying a strategy by actually following it in the present as long as it works, then changing to a different strategy, usually one that has been doing well recently. This, of course, is just back testing in disguise. If you select strategies for forward testing on the basis of how well they have performed in the past, you are back testing. Worse, after your forward test is complete, it has become a back test for any decisions about the future. That s because you can only evaluate a forward test after you know what its performance was. Notice the past tense! All this said, it is still true that the past is the only thing we have to help us predict the future. Optimization and back testing using historical data can be useful both in improving the chances of success (not by very much) and even more importantly to provide an investor with realistic expectations for attempts to produce future returns based on past performance. A Sample Program for Testing Risk Control Strategies This program is presently capable of testing only those strategies based purely on price movements. The data it uses contains volume information, but this is not used in the present version, although it could easily be added by someone familiar with Visual Basic for Applications or VBA. This is Microsoft s standard language for writing programs for their spreadsheet program Excel. It was written using Excel 2

3 2000, and is generally compatible with later versions. It has been tested with Excel 2003 and 2007, and seems to work, although complete compatibility is not guaranteed. It has not been tested with any version of Excel for the Mac, but there is a good chance it will work there as well. The program is being released under the GNU general public license, the short version of which is shown above. The full text of the license is included with the program on the worksheet labeled GNU GPL. Running the Program This is not a stand-alone program. It must be run by opening the supplied spreadsheet (Risk.xls) using Microsoft Excel 2000 or above. If you have Microsoft Office 2000 or later on your computer, you probably have an appropriate version of Excel. Just open the file in Excel. It is assumed that you have a basic familiarity with Excel. Programming experience is not necessary to use the program. Basic Tutorial This tutorial will illustrate some uses, limits and pitfalls of optimization and back testing. The Input Worksheet 3

4 The input worksheet shown above is the location where all of the control functions for the program are located. It can be opened by clicking on the Input tab at the bottom of the workbook. The worksheet is protected (no password is needed to remove protection) so the only place you can type is in the gray areas. In addition there are three green buttons. One button downloads data, one starts the optimization program to find the best settings for a particular strategy, and the third back-tests a specific strategy with specific settings over one or more periods in the past. Downloading Data Using the Input Worksheet First type a list of tickers in the Ticker column at the top left. Sectors or any other identifying label can be typed in the Sector column, but this is not required. If such labels are included they will be displayed on the worksheet to which the data is downloaded. The program will accept up to 30 tickers, but the downloads and optimizations may be slow if more than ten or so are used. Second, type in the date range you wish to download in the From Date and To Date boxes, and either d, w, or m for daily (market days only there are about 252 market days per year), weekly or monthly adjusted close data in the download. Yahoo Finance adjusted close data is adjusted for splits and assumes reinvestment of dividends. Third, press the Green Download Data button. The Destination Worksheet input box appears. The default worksheet is labeled Data, or the general purpose worksheet Scratch can be used. The other normally unprotected worksheet is OptTables, but it should generally not be used for downloads. Trying to use any protected worksheet such as Input, OptResults, BTResults and GNUGPL will generate an error message. Of course you can use Excel functions add other worksheets to keep different data sets if you choose. WARNING ALL PREVIOUS CONTENTS OF ANY UNPROTECTED WORKSHEET TO WHICH DATA IS DOWNLOADED WILL BE ERASED! A good way is to use the default Data sheet for downloads, but first manually copy any data you want to save to another sheet you ve manually created for the purpose. Data Details The program works with publicly available historical data from Yahoo Finance website ( This data can be downloaded manually from the website, but, as just discussed, the program will automatically download the data it needs. Only normalized adjusted close price data are saved at present. The data are normalized to a price of 1 at the earliest download date. The program is intended to be an example program intended to test strategies based on closing price movements alone, so volume, open, close etc. are not saved in this version. It would not be difficult data for those of you with programming experience to save the other data, and very simple to save raw price data (just comment out the normalization routine). 4

5 The program corrects for missing or extra days sometimes found in the online data, so all tickers cover exactly the same input dates. If this occur or tickers can t be found an information message appears in the Messages column. The data must be placed on the input worksheet in a very specific format. Changes to that format will produce incorrect results, so be careful if you manually add data to the worksheet. However, if you keep the format intact, you can import data from other sources. Here s what a short (6 market day) download looks like: ETF ETF ETF ETF Date IWM QQQQ SPY DIA 6/5/ /2/ /1/ /31/ /30/ /26/ Take a look at the Data worksheet after a download. One advantage of normalized data is that is that you can see at a glance the percentage increase or decrease over time. QQQQ (a NASDAQ ETF) went up over 20% between May 26 and June 5, Using Mechanical Buy and Sell Techniques for Protection We often hear that wise investors never buy or sell a security on price movements alone. I agree with that philosophy, and in both my professional and personal investing I use a wide range of fundamental data, news, economic data, and historical information. However, once one is past the absolutely critical task of identifying good and bad investments, it is also necessary to deal with the fact that most price movements in a stock are caused by factors other than the business prospects of the company it represents. The fact is that market psychology can cause healthy companies to drop drastically, often for years, or cause obviously worthless businesses to be valued at totally unrealistic high prices, often for years. This fact is the reason that so-called value investing has consistently been the most successful long-term approach to investing. John Maynard Keynes once remarked that, It makes no sense to buy a stock that you believe the market is valuing at twice its true value, if you simultaneously believe that the marker will value it at half its current value two months hence. I would add that if you don t know what the market will do, or if you re not sure of how accurate your value estimate is, or if you are simply willing to admit that unforeseen things happen, then you should protect yourself against the possibility that what you thought was an excellent and safe investment might go wrong. Mechanical buy and sell methods are one way to do that. Diversification and protective options are others. 5

6 Program Capabilities This program deals with a few examples of price-driven buy and sell strategies. These are: 1. Buy if the price has change a given percentage (either up or down) in a given time. Examples: The price has dropped 5% or more in the last 5 market days (buy on weakness). The price has increased 5% or more in the last 5 market days (buy on strength). 2. Sell if the price drops a fixed percentage below your purchase price (fixed price stop). Example: You bought at $100 and you sell if the closing price drops 5% below your purchase price or at $ Sell if the price drops a fixed percentage below the highest of the purchase price or the highest subsequent high (trailing stop). Example of a 5% trailing stop based on the day s close: You buy at $100, the stop is at $95; the next day s close is $105, the stop is at $99.75; the next day s close is at $101, the stop remains at $99.75; the next day s close is at $110, the stop goes to $104.50, etc. 4. Sell when the closing price reaches a fixed percentage above the purchase price (protect profits). Example: You buy at $100, and will sell when you have a profit of 10%, so you sell when it closes above $110. IF BOTH BUY AND SELL SIGNALS ARE PRESENT AT THE SAME TIME, THE PROGRAM DOES NOTHING. Because we are limited in this program to adjusted closing prices these are what are used in all these cases. Volume could easily be added, and other price data adjusted by the program. There are may other price-driven buy and sell strategies that could be easily incorporated. One example uses moving averages to identify significant market moves, another uses standard deviations to set trailing stops. A programmer could easily implement new strategies such as these. Running the Optimizer The key questions in using these the strategies described above are, What percentage price change do I use with a given stop or buy strategy for the best performance? and Which of these strategies is best? The Optimize button deals with the first question directly, and used with different strategies can answer the second. Date Range and Sampling Interval: The optimizer deals with the downloaded data only in the format that is downloaded, using all the downloaded data. The easiest way to change the date range or sampling interval (market day, week, or month) that is to be optimized is to download new data as required. You can manually adjust the data as long as the required format is preserved. 6

7 The optimizer uses dates beginning with row 3 and continues until a blank date is encountered. It uses tickers starting with Column B and continues until a blank ticker is encountered. Using a download of daily data over a given time span, the data can be manually set to a shorter time span as follows, or just download a shorter time span. Move the data columns to the right. Because of the intervening blank cells the optimizer won t see them. Copy what you want to the top left, and then run the optimizer: 7

8 Keep in mind that back testing is statistical analysis, and that you need enough data for meaningful results. Therefore, you should not trust data with fewer than about 100 dates in it. The above examples have only a few dates in them so they will fit on the page, but you should not use downloads that short. Even if you have enough data, the results of using the optimizer are only valid if the past data used is representative of what will occur in the future, something that can never be known or guaranteed. Other Settings: Once you have your data ready, click the green Optimize button on the Input worksheet. In the box that appears, if the default worksheet Data is not correct, type the name of the worksheet on which the data is stored, then click OK. The Optimization Methods dialog box appears: In this example we have downloaded over 2000 lines of daily data covering the period May 26, 2000 through the present. This is plenty of data from a statistical point of view, and covers the tech bubble bust and recovery, 9/11, and the recent boom, bust and partial recovery. Anything that works consistently over all that is likely to be pretty robust. 8

9 The tickers IWM, QQQQ, SPY and DIA are popular high volume exchange traded funds that track the Russell 2000 small cap index, the NASDAQ, the S&P 500, and the Dow Industrials respectively. They are a good cross section of U.S. stocks. There is only one buy strategy presently implemented. (For those VBA programmers among you with the inclination, it s pretty easy to add others.) Therefore, it doesn t need to be selected. It looks for stocks with a price that has changed by a certain amount over a specified period of time. In the example the specified period of time is 5 periods, which is 5 market days because the sampling period of the data is one market day. This and other numbers in the Fixed Value boxes are not varied in the optimization process. We have specified that the percentage by which the price must change be varied from a Low of 10% to a High of +10%, in steps of 1% We have selected a sell strategy using trailing stops, which sell the next day after the price drops by a certain percentage below the highest of purchase price or any subsequent closing high. Because it uses a price drop, it must be entered as a negative percentage. There are no fixed value numbers to be entered. We have specified that the optimizer vary the trailing stop percentage in steps of 1% from 0% (that is, any down day after buying) to a -20% down day. The above numbers and strategies are the program defaults, and cover the majority of market situations. Trailing stops have the nice features of protecting you from large losses and also protecting your profits once the price has appreciated above the stop percentage. They do this without limiting your profits if the price continues to rise. There are also two other sell strategies that can be used. These include fixed stops that sell when the price drops a certain percentage below the purchase price, and a protect profits strategy that sells at a certain percentage above the purchase price. We will discuss both of these in more detail later. Processing Time Required The optimization can take quite a while if you have a lot of data and tickers. This example took about 18 minutes to run on a 3 GHz Pentium 4 with 1 Gigabyte of memory running Windows XP. The optimization is an exhaustive search of each ticker, so 20 tickers would take about five times as long as four, all else being equal. Don t search percentages with a step less than 1%. Going to say 0.1% over the same range for both buys and sells takes about 100 times as long, and probably won t tell you anything useful. Any profit increase that depends on a 0.1% change in a stop value is an illusion. The future doesn t duplicate the past that closely. On the other hand, using a 2% or even 5% may work just fine, and takes a lot less time to process. 10% is too coarse. It doesn t happen too often in the stock market over a one-day interval. If you are using weekly or monthly data you can increase the range and step size of the data accordingly. Once you are happy with the settings in the Optimization Methods box, click on the Enter button. 9

10 Optimizer Results Sell Method Buy Method Start Date Stop Date Period TS 5 mkt day % chng 5/26/00 11/25/09 mkt day Ticker IWM QQQQ SPY DIA Std Dev in Period 1.67% 2.08% 1.40% 1.33% Mean Return 0.03% 0.00% 0.01% 0.02% Buy & Hold Return 44.46% % -5.52% 22.88% Max Buy & Hold Draw Down % % % % Max Draw Down Date 3/9/ /9/2002 3/9/2009 3/9/2009 Optimum Buy -9.00% -5.00% 8.00% 7.00% Optimum Sell % 0.00% -9.00% -9.00% Optimum Return % % % % Optimum Drawdown % % -9.74% % Total Trades with Opt Strategy Winning Sales Losing Sales Transaction Dates and Types IWM QQQQ SPY DIA 09/17/01 B 07/24/00 B 07/29/02 B 04/10/01 B 07/02/02 SW 07/26/00 SW 09/17/02 SL 07/06/01 SW 07/23/02 B 07/27/00 B 10/15/02 B 07/29/02 B 10/07/02 SL 08/03/00 SL 01/27/03 SL 09/17/02 SL 10/09/02 B 09/08/00 B 03/18/03 B 10/15/02 B 01/08/08 SW 09/13/00 SL 08/15/07 SW 01/27/03 SL 09/29/08 B 09/14/00 B 10/31/08 B 03/17/03 B 10/14/08 SL 09/15/00 SL 11/06/08 SL 11/21/07 SW 10/24/08 B 09/21/00 B 11/26/08 B 10/31/08 B 11/18/08 SL 09/25/00 SW 01/14/09 SL 11/06/08 SL 11/19/08 B 09/27/00 B 03/12/09 B 11/26/08 B 02/18/09 SW 09/29/00 SL 11/25/09 SW 01/20/09 SL 02/23/09 B 10/03/00 B 03/12/09 B 11/25/09 SW 10/09/00 SL 11/25/09 SW 10/10/00 B The Optimizer Results worksheet is protected from user change (no password is needed to remove protection). The results should be copied to the Scratch worksheet or another worksheet you add if you want to work with them for further analysis. The top two lines show the strategies tested and the start date, stop date and sampling period of the data used for the optimization. The next lines show the tickers used, and the basic statistics of their price movements as follows: 1. The standard deviation of the returns over one sample period (in this case one market day) 2. The average return per sample period (for IWM 0.03%) 3. The buy-and-hold return for an investor that bought on the first days close (5/26/00) and sold at the last days close (11/25/09), (for IWM 44.46%) 4. The maximum drawdown (price drop from a previous high during the period analyzed), (for IWM 58.86%) 5. The date this drawdown occurred (for IWM 3/9/2009) 6. The value at the optimum for the buy signal (for IWM a 9% drop over 5 market days) 7. The value at the optimum found for the sell strategy (for IWM and a trailing stop, a 17% drop) 8. The total return at the optimum (for IWM %) 9. The maximum drawdown at the optimum (for IWM %) 10

11 10. The total number of trades in the optimum strategy, both buys and sells counted (for IWM 14) 11. The number of winning sales in the optimum strategy (for IWM 4) 12. The number of losing sales in the optimum strategy (for IWM 3) Below this data is a list for each ticker of the date of each trade the type of each trade (buy B or sell S) and the outcome of each sale (win W or lose L). Breaking even is a win. It is assumed that all securities held are sold at the end of the optimization period, so the last transaction is always a sale, even if no stops are triggered. This allows a fair evaluation of actual money earned using the strategy. (There were 192 transaction in QQQQ, so the whole table is not shown here.) Do these results mean that if we buy IWM today and thereafter follow the optimum strategy as found, that we could expect to beat a buy-and-hold strategy by 140% or over 4 to 1 over the next 10 years? No! To see why, we need to look at the other output sheet of the optimizer, OptTables, shown below: There are tables like this for each ticker. They show the returns for every combination of buy and sell values tested. They show the return achieved for the optimum in bright green, and other winning values in dark green. White boxes are losses. Interpreting Results The thing to notice is that the optimum found is sitting near the edge of small area of green, and that if both buy and sell percentages were different by just 1% (buy on a 10% drop, sell with a 16% trailing stop) that 185% gain would have been a loss of -22%. That strongly suggests that the optimum might be due to hindsight that matched exactly one or more of the major market moves. These moves are unlikely to be matched in the future to this degree of precision. In addition the large and robust patch of green further right has return values around the buy-and-hold number of about 44% for a wide range of values. We can see that they involve buying on a small increase in price and selling on a large drop. This is suspiciously like a buy-and-hold strategy. 11

12 Let s go back and look at the trades at the optimum. Here s a chart of IWM from 5/26/00 to the present, which we made by plotting the downloaded data and the trades made on the Scratch worksheet: /24/ /6/1999 4/19/2001 9/1/2002 1/14/2004 5/28/2005 2/22/ /10/2006 7/6/ /18/2010 : Using the optimum strategy of buy on a 9% loss in 5 market days and sell on a 19% trailing stop, we were in the market where the line is white and out where it s dark blue. It did a good job of keeping us in sustained bull markets, and kept us out of downward drifting markets. It did a poorer job on sharp declines, but still kept us out for short periods of time, which limited our losses during those times. The trades and amounts made were calculated on the Scratch worksheet. They were: From To % Gain 9/17/2001 7/2/ % 7/23/ /7/ % 10/9/2002 1/8/ % 9/29/ /14/ % 10/24/ /18/ % 11/19/2008 2/18/ % 2/23/ /25/ % This looks general enough to try a second optimization with the value range extended to move the optimum closer to the center of the charts. That might find a better value with more room for future changes in volatility. 12

13 Here s the result: IWM % % % % % % % % % % % -9.00% -8.00% -7.00% -6.00% % % % % % % % % % % % % % % % % % % % % % There are no better optimums in the vicinity of the one found, but a somewhat safer bet than the actual optimum might be a buy at 8% in 5 days and a sale with a trailing stop of -18% because there s more margin for error, and the optimum can and will move around in the future. Basically, we have found that at least over the last ten years, for this security, you have to maintain a trailing stop of about 16 to 21% to avoid selling out during temporary downturns in a bull market, and that buying at about 8 or 9% in 5 days will be a good way to buy in the dips. However, the basic strategy is not very effective at avoiding sharp, extended sell-offs. Still, it has the potential to outperform buy-and-hold. Back Testing For this let s use the Dow Industrials. Data is available for the index itself as far back as October 1, We ll use that data, and do a back test using daily data from 1949 to the present. Why 1949? In 1947 the Fed became a major influence. It was the modern era. The most valid back tests are based on data in the past, so we will first do an optimization on ^DJI, Yahoo s symbol for the Dow using data from 1949 to Then we ll back test the resulting optimum. This was near the start of the great post-war (the last real war the US fought World War II) boom. The war got us out of the depression, we won and the men came home. The GI bill and a rush to convert factories from wartime to peacetime products was on, and everyone had a job. The Fed was seriously trying to control the money supply and another great depression seemed very unlikely. It was a good time to invest. Here s what you would have found if you ran this analysis in 1959: 13

14 Ticker ^DJI Std Dev in Period 0.67% Mean Return 0.05% Buy & Hold Return % Max Buy & Hold Draw Down % Max Draw Down Date 10/22/1957 Optimum Buy -2.00% Optimum Sell % Optimum Return % Optimum Drawdown % Total Trades with Opt Strategy 4 Winning Sales 2 Losing Sales 0 Transaction Dates and Types Sell Method Buy Method Start Date Stop Date Period TS 5 mkt day % chng 11/25/49 11/25/59 mkt day ^DJI 01/13/50 B 10/14/57 SW 10/21/57 B 11/25/59 SW The optimum barely squeaks past simple buy-and-hold. There are just four transactions in 10 years. Here s a chart of the market showing white when we were in, and dark blue when we were out: /1/1949 5/16/1950 9/28/1951 2/9/ /15/1959 8/2/1958 3/20/ /6/1955 6/24/

15 If you look very closely at that big drop in 1957, you ll see a tiny dark speck in the graph. Here s a closer look at it: /1/1957 4/20/1957 6/9/1957 7/29/1957 9/17/ /6/ /26/1957 2/14/1958 Here s the table: ^DJI % -9.00% -8.00% -7.00% -6.00% -5.00% -4.00% -3.00% -2.00% -1.00% 0.00% 1.00% 2.00% 3.00% 4.00% 5.00% 6.00% 7.00% % % % % % % % % % % % % % % % % % % % % % Now, that is as close to a guaranteed profit as anything you ll ever see! In a market like that, you could lose, but you would need really bad luck. The optimum is in the middle of a sea of green. The zeros on both sides are break-evens because during this time there were no 5-market-day periods when the Dow went up as much as 6% or down 8%. Even so, as the graph and results show, it helped a little to have an exit trigger to get out when things went bad. That probably didn t seem like such a dumb idea to people fresh out of 15

16 the depression. The numbers don t seem unreasonable. After all, a 14% trailing stop is big enough that it would probably be triggered by a serious problem. But after a while, people forget about serious problems they haven t seen for a while. First we ll do a back test using all the data, and then we ll do a back test in 10-year increments starting with 11/25/1959 running up to the present. We ll use the optimum values found in this test. These are: (1) buy on a 2% drop in 5 days, and (2) sell on a 14% trailing stop. (As I write this, I have no idea what the results will be.) First download data running from11/25/1959 to 11/25/2009. Click on the green Backtest button on the Input worksheet, and then type in the name of the worksheet where the data is, if it s not the default Data. The Backtest dialog box appears. Type in 2.0 in the % Change box under Buy Method and type 5 in the In N Periods box. Then enter 14.0 in the % Change box under Sell Method. Make sure Trailing Stop (%) is selected in Sell Method. For the first test leave the defaults in the Backtest Inputs boxes. These are the words Earliest for Earliest End Date, Latest for Latest End Date and All for Test Length. The dialog box should look like this: Now, click on Enter. 16

17 The results from the BTResults worksheet are as follows: Ticker ^DJI Std Dev in Period 1.01% Mean Return 0.03% Buy & Hold Return % Max Buy & Hold Draw Down % Max Draw Down Date 3/9/2009 Backtest Buy -2.00% Backtest Sell % Backtest Return % Backtest Drawdown % Total Trades with Opt Strategy 54 Winning Sales 14 Losing Sales 13 Sell Criterion Buy Criterion Start Date End Date Data Interval Backtest Period -14% TS 5 mkt day -2% chng 11/25/ /25/2009 mkt day mkt days Transactions by Ticker ^DJI 01/12/60 B 06/01/62 SL 06/04/62 B 08/03/66 SW 08/16/66 B 07/18/69 SW 07/24/69 B 04/30/70 SL 05/04/70 B Because of the use of reasonable mechanical stops and buy rules, the net return over 50 years was about 18% higher than buy-and-hold. $10,000 invested in the Dow in 1959 would have returned $160,719 by now with buy-and-hold. With the stops and buy criteria the return would have been $189,394. There were only 54 trades, both buys and sells, in 50 years. Now we ll try splitting the 50 years into 10 separate intervals. This will not necessarily add up to the same returns as what we have just seen, because when you sell and buy again to start over, you ve re-set the previous high in the trailing stop to whatever your new purchase price is. It is, however, a valid technique to see how the strategy would have performed in a range of different conditions. In the results above the back test period is given as market days. To split the 50 years into 10-year segments divide that number by 5 and round down / 5 = = 2517 market days. Use the same data and open the Backtest dialog box. Change the Test Length box from All to Leave everything else the same. The full results are printed just like the one above, one for each ten-year interval, left-toright on the BTResults worksheet. WARNING THERE ARE ONLY 256 COLUMNS ON A WORKSHEET. Each result requires at least 8 columns, more if multiple tickers are used. If I had used 1 year intervals, that would have been 400 columns. DON T EXCEED 256 COLUMNS! The back test will start at the present day and work back in 2517-day chunks until it has less than 2517 days of previous data. Then it stops. Now we re ready, so click on Enter. 17

18 Rather than show the full results for each interval in this manual, here is a manually edited summary: Start Date End Date 11/23/ /25/2009 Buy & Hold Return -4.83% Backtest Return -5.25% Max Buy & Hold Draw Down % Backtest Drawdown % Start Date End Date 12/7/ /23/1999 Buy & Hold Return % Backtest Return % Max Buy & Hold Draw Down % Backtest Drawdown % Start Date End Date 12/24/ /7/1989 Buy & Hold Return % Backtest Return % Max Buy & Hold Draw Down % Backtest Drawdown % Start Date End Date 1/8/ /24/1979 Buy & Hold Return 4.62% Backtest Return 8.61% Max Buy & Hold Draw Down % Backtest Drawdown % Start Date End Date 11/27/ /24/1979 Buy & Hold Return 22.92% Backtest Return 30.71% Max Buy & Hold Draw Down % Backtest Drawdown % In every interval but the most recent the stops improved returns over buy-and-hold. During the period 11/23/1999 to 11/25/2009 Buy-and-hold lost 4.83% while the protective strategy lost 5.25%. However, the strategy had a drawdown roughly half that of buy-and-hold. In all other periods, it outperformed buy-and-hold and equaled or bettered the drawdown. Remember it was developed in a strong bull market in Other Sell Strategies Using a sell strategy of selling at or above a given profit percentage (PP or Protect Profits in the program) can be a safe, but relatively low return method of investing. Rather than discuss it here, the program is posted with a worked example in it, including a manual Excel analysis on the Scratch worksheet. You might find it interesting. I personally do not like fixed stops, which sell at a fixed percentage below the purchase price. They do no better job protecting you from losses than a trailing stop and don t preserve profits like a trailing stop will. The major problem is that they guarantee a loss on every sale. I have known people using fixed-price stops who lost huge percentages of their funds in a sideways market 5% at a time. The strategy is available. Try it, you won t like it at least as a routine part of a long term strategy! 18

19 Final Thoughts There are no guarantees that any strategy will work. For example, after 9/11 the markets were closed for some time, so you couldn t have made your trades. In a few thankfully rare cases markets have fallen as much as 20% in a single day, something that a system with once-a-day data wouldn t catch until too late. Changes in volatility can drastically change performance, especially when fees are included, as they should be. This program does not account for inflation or transaction costs, and assumes reinvestment of dividends, which tends to overstate real returns. All of this said, I have found that the criticism often leveled at such investing techniques is not really valid. You often hear that when you get out of rapidly falling markets, you will be afraid to get back in until the market is higher than when you left. That isn t true with the example strategy above if you actually follow it. All it requires to get back in after you sell is one week with more than a 2% drop in closing price (and no simultaneous sell signal). That s why the strategy is in the market most of the time. But it keeps you out of it enough to outperform buy-and-hold and does it with lower drawdowns. Have fun. Experiment. Try weekly or monthly data instead of daily. This might be better for investors who only want to think about their portfolios once a week or once a month. Try different time periods over which the market must rise or fall for a buy signal. If you run into problems with computer compatibility, crashes, etc. please let me know. I might or might not be able to fix it, but at least I can warn people about it. I can be reached at smithson@anava.com. 19