I only just now have time again to discuss this topic. Does anyone still care about testing on adjusted data verses authentic data?
Yes. I thought I had this figured out but having some objective feedback wouldn't hurt at this point. In backtesting, I use the adjusted close for my technical analysis and the actual next open for my transaction price. The thing I do to reconcile the two is to multiply my market position (number of shares owned or owed) by the following ratio: the product of today's adjusted close and yesterday's actual close divided by the product of today's actual close and yesterday's adjusted close This ratio is typically one except on days of a split or a dividend payout. I've been assuming that this is the only correction needed. Constructive comments are welcomed.
To be honest, your ratio sounds very peculiar. Using the opening price as your transaction price is pretty good. The reason for using a ratio of number of shares would be to simulate reinvestment. I am assuming that your "adjusted close" is a back-adjusted series. In that case, the yesterday's adjusted close would be very close to the actual close. Is your back-adjustment being done proportionally? Given my poor understanding. Let's look at a hypothetical stock that earns 1% per day. It had a 5% stock dividend and a 2:1 split. You can see that using the actual number of shares (reinvestment assumed) and my adjusted series show the same return on investment. The Simulation numbers shown reflect my poor understanding. The Simulation numbers, in particular, show this as being a terrible trade which it wouldn't be. Code: Date Act Pri Event Act # Act Bal Daily ROI Adj Pri Daily ROI kut2k2 Rat Sim # Sim P/L 02/01/11 100.00 1000 100000.00 47.50 02/02/11 101.00 1000 101000.00 0.01 47.98 0.01 1.01 500 0.00 02/03/11 102.01 1000 102010.00 0.01 48.45 0.01 1.01 505 510.05 02/04/11 103.03 1000 103030.10 0.01 48.94 0.01 1.01 510 1035.45 02/05/11 104.06 1000 104060.40 0.01 49.43 0.01 1.01 515 1576.57 02/06/11 105.10 1000 105101.01 0.01 49.92 0.01 1.01 520 2133.76 02/07/11 100.84 5% Dividend 1050 105886.64 0.01 50.42 0.01 0.96190476 500 -77.87 02/08/11 101.85 1050 106945.50 0.01 50.93 0.01 1.01 505 431.11 02/09/11 102.87 1050 108014.96 0.01 51.44 0.01 1.01 511 955.42 02/10/11 103.90 1050 109095.11 0.01 51.95 0.01 1.01 516 1495.43 02/11/11 104.94 1050 110186.06 0.01 52.47 0.01 1.01 521 2051.49 02/12/11 52.99 2:1 Split 2100 111287.92 0.01 52.99 0.01 0.505 263 -12625.77 02/13/11 53.52 2100 112400.80 0.01 53.52 0.01 1.01 266 -12611.25 02/14/11 54.06 2100 113524.80 0.01 54.06 0.01 1.01 268 -12593.76 02/15/11 54.60 2100 114660.05 0.01 54.60 0.01 1.01 271 -12573.21 02/16/11 55.15 2100 115806.65 0.01 55.15 0.01 1.01 274 -12549.51 02/17/11 55.70 2100 116964.72 0.01 55.70 0.01 1.01 276 -12522.57 If you can nail down what you are really trying to measure/simulate, then there will be only one correct way to adjust your data to make that metric invariant to the adjustment.
All I'm doing is using the adjusted price data to account for splits and dividends. Perhaps I should have used the term "correction factor" instead of ratio. Let's look at the split. The day before, the adjusted price (presumably the adjusted close) is 52.47 and the actual close is 104.94. The day of the split, the adjusted close and the actual close are both equal to 52.99. So the correction factor (the product of today's adjusted close and yesterday's actual close divided by the product of today's actual close and yesterday's adjusted close) becomes (52.99x104.94)/(52.99x52.47) = 2.00, exactly as expected. Multiply your current market shares by 2 to get the new number of market shares. Where on earth did you get 0.505 from?
Two things: 1. Technical analysis using the adjusted close can be highly misleading at times. I already posted a link to a specific example for the SPY where a major double top dissappears after adjusting the data. 2. Backtesting using series adjusted for dividends can generate misleading results. This is a link to Part 2 of the first link with a specific example: http://www.priceactionlab.com/Blog/...ting-part-ii-more-on-close-vs-adjusted-close/
I totally disagree with your conclusions. The whole point of using adjusted data in TA is that you've already accounted for dividends and splits. No need to go back and look at each dividend and split when someone has already done that for you by providing adjusted data, once you've verified that the adjusted data can be trusted (the same issue you have with the raw data.) The example of the split is the most illustrative. Once a stock splits, the raw price chart looks like a large drop has taken place when actually no loss in stock value has taken place. The adjusted data automatically accounts for this and the chart of it reflects this. Seems to me like you're advocating a whole lot of extra work for no real payoff. If the adjusted data is good enough for computing the buy-and-hold result, then it's good enough for backtesting a trading strategy as well.
He was talking about dividend adjustmets, not splits. Of course you have to adjust for splits in stocks and rollovers in futures. His argument was that adjustments for dividends distort the technical picture. This is obvious and I fully agree. You keep on fighting it won't make you right. Just look at that annimation on the example he gave: http://www.priceactionlab.com/Blog/wp-content/uploads/2011/03/D9HGlb1.gif Does the dividend adjusted chart look right to you for the purpose of technical analysis? If it does, you just go ahead. This is a free country. Do as you like. But don't come here and tell us that you missed a triple-top in S&P 500.
I have yet to see any adjusted price data that didn't account for both dividends and splits. The whole example is contrived nonsense. And I couldn't give a flying fig about triple tops, quadruple bottoms, flags, pennants and wedgies. Most if not all of that cloud pattern stuff is no less voodoo than your fibonacci fetish. I do real math, not voodoo math.
tim888, kut2k2, intradaybill: Don't be so cut-and-dry. If your hypothesis is that 10,000 is an important support/resistance point, then there is no point in basing trade entry/exit decision based on data adjusted to a level different from that which the crowd is seeing. If you want to know what kind of return your buy-and-hold system would have had over the last 30 years, then there is no point in neglecting the compounding effects of dividends. Just as there benefit in looking at multiple timeframes, there are benefits to looking at the same data with different adjustments. The case of equity adjustments is trivial when compared with futures, fixed-income, structured interest rate products, and options. The goal, however, is the same. Apply the right data to each task to get the truest answer without adding unnecessary complexity.
This is what you stated: The point is that no person in the right state of mind would use dividend adjusted data for backtesting, in general, because at the time of the backtest, these data were not adjusted. Gains for dividends can be trivially added after the backtest. I know it is hard to realize that all this time you were doing the wrong thing. Instead of being aggreessive and trying to mislead others, thank people for showing you the right way. Aggression will get you nowhere. Admitting mistakes is a sign of intelligence...