Yahoo Finance has an interesting article on return after n-consecutive days. http://biz.yahoo.com/tm/070126/15384.html Seems like this is a strong case for mean reversion methods over trend following methods. Of course if one was able to predict that n-consective days consistently led to (n+m)-consecutive days, then of course trend following would be a good method. I'll look forward to the next article where they filter the consecutive days with a 200-day MA, and then examine returns.
I am writing a computer program to explore the TradingMarkets Research observations. The code that I have looks good but there might be bugs and I need to validate the program. I am testing 5 consecutive daily sessions lower closing prices for an entry signal, exit when price is greater than the 50 session moving average. This test uses 5 % heat and stops losses at 10 % (no losses are stopped in this simulation). System trades long only. Compensations are made for commission and slippage. Data is 13.97 years of Standard and Poors 500 index tracking stock, symbol SPY. Number of trades 14 Total profit $ 5979 Profit after subtracting $ 10.00 commission, slippage per transaction: $ 5699 Heat is 5.00 per cent of equity. Greatest drawdown is 0.0172 (1.72 per cent). Cumulative Annual Growth Rate (CAGR) is 0.41 per cent. CAGR / Drawdown is 0.24 Instantaneously Compounding Annual Growth Rate (ICAGR) is 0.40 per cent. Annually Compounding Annual Growth Rate (ACAGR) is 0.40 per cent. Information Ratio is 0.54 Initial capital is $ 100000
markets change volatility is of course cyclical and the market (which is everybody's aggregate decisions) necessarily adapts. the 90's were a higher volatility market, and also a market where mean reversion was not as successful this aint yer daddy;s bull market it is a great means reversion market now.
when i ran your strategy it worked over the last few years. History of the portfolio : -------------------------- Long position (0) on spy 2004-07-20 Buy 100 at 110.5300 2004-08-26 Sell 100 at 110.9600 Long position (1) on spy 2004-08-09 Buy 100 at 107.0200 2004-08-26 Sell 100 at 110.9600 Long position (2) on spy 2005-01-05 Buy 100 at 118.7400 2005-01-13 Sell 100 at 118.6400 Long position (3) on spy 2005-01-06 Buy 100 at 118.4400 2005-01-13 Sell 100 at 118.6400 Long position (4) on spy 2005-01-25 Buy 100 at 116.9100 2005-02-02 Sell 100 at 119.0600 Long position (5) on spy 2005-09-16 Buy 100 at 123.2800 2005-10-03 Sell 100 at 122.9600 Long position (6) on spy 2005-10-07 Buy 100 at 119.6700 2005-11-03 Sell 100 at 122.1500 Long position (7) on spy 2005-10-14 Buy 100 at 118.1000 2005-11-03 Sell 100 at 122.1500 Long position (8) on spy 2006-01-19 Buy 100 at 128.1300 2006-01-24 Sell 100 at 126.6800 Long position (9) on spy 2006-03-08 Buy 100 at 127.6500 2006-03-13 Sell 100 at 128.8300 Long position (10) on spy 2006-06-09 Buy 100 at 126.4200 2006-07-05 Sell 100 at 127.2900 Long position (11) on spy 2006-06-12 Buy 100 at 125.8500 2006-07-05 Sell 100 at 127.2900 Long position (12) on spy 2006-06-13 Buy 100 at 123.7400 2006-07-05 Sell 100 at 127.2900 Long position (13) on spy 2006-06-14 Buy 100 at 122.8800 2006-07-05 Sell 100 at 127.2900 Long position (14) on spy 2006-07-18 Buy 100 at 123.7100 2006-07-25 Sell 100 at 126.0300 Long position (15) on spy 2006-08-10 Buy 100 at 126.5800 2007-01-26 Sell 100 at 142.0900 Long position (16) on spy 2006-08-25 Buy 100 at 129.6400 2007-01-26 Sell 100 at 142.0900 Long position (17) on spy 2006-11-02 Buy 100 at 136.4600 2007-01-26 Sell 100 at 142.0900 Long position (18) on spy 2006-11-03 Buy 100 at 137.2000 2007-01-26 Sell 100 at 142.0900 Long position (19) on spy 2006-11-06 Buy 100 at 136.9800 2007-01-26 Sell 100 at 142.0900 ## Global analysis (full portfolio always invested) Analysis of the portfolio (2004-05-11 / 2007-01-26) : ----------------------------------------------------- Performance : 46.0% ( 14.9%) Buy & Hold : 32.0% ( 10.7%) () => by year MaxDrawDown : 2.8% B&H MaxDrawDown : 7.6% Best performance : 46.0% Worst performance : -0.5% Net gain : 4598.11 Gross gain : 6874.00 Trades statistics : Number of trades : 20 Trades/Year : 7.38 Number of gains : 14 Number of losses : 6 Win. ratio : 70.0% Max consec. win : 9 Max consec. loss : 2 Expectancy : 0.02 Average gain : 2.94% Average loss : -0.93% Avg. perf : 1.91% Biggest gain : 11.26% Biggest loss : -2.04% Profit fac : 3.16 Sum of gains : 5280.53 Sum of losses : -682.42 Risk of ruin : 0.0%
Same SPY data but traded with Bollinger Band trend following system. Parameters are 400 days moving average, 0.5 standard deviations, 5 % heat. Same time period (13.97 years from 1993-01-29 to 2007-01-26.) Allowance made for commission and slippage. Number of trades 10 Total profit $ 102448 Profit after subtracting $ 10.00 commission, slippage per transaction: $ 102248 Heat is 5.00 per cent of equity. Greatest drawdown is 0.0228 (2.28 per cent). Cumulative Annual Growth Rate (CAGR) is 7.32 per cent. CAGR / Drawdown is 3.21 Instantaneously Compounding Annual Growth Rate (ICAGR) is 5.04 per cent. Annually Compounding Annual Growth Rate (ACAGR) is 5.17 per cent. Information Ratio is 0.45 Initial capital is $ 100000 Long trades only.
How much initial capital did you use in the simulation? Did you intend to have overlapping trades? For example, buy on 2004-07-20 then buy again at 2004-08-09 and sell both 2004-08-26. Nothing wrong with that. Might improve the method.
Very interesting subject. Reversion to mean probably changes a lot when selecting diff ( longer) time period. Stock with high and steady PEG ratio may never return to the same price , especially in overall PE's expansion environment.
I can examine different time periods and have some hourly and minute data to test. Right now I am more concerned about validating my code and eliminating programming bugs.
Here is a copy of the SPY data that I am using for the test. I downloaded it from yahoo.com today. If we are testing this trading method as a group we might use the same data.
HNS , article compares results of XYZ change vs benchmark's ( a la pair trading bases) ; you are testing totally diff strategy , no ?