Before testing many years of data, test your own computer backtest by manually testing the last 10 days & compare to your computer backtest. Sorry, I think too many computer backtests aren't tested very accurately.
Just looking at weekly data 2018 was not profitable lost 213 ES points, 2019 so far made 92 points using 1 lot. looks like works better with smaller weekly ranges. I will backtest it but any suggestions as far as stops? low of opex week? fixed stop? test with ES or SPY? anything else? this backtest will be done as strategy with optimization of inputs.
That's cool that you're willing to do this. However, I can say with near certainty that any strategy this simple either (1) never worked or (2) may have been profitable at one time but was arbitraged out of any profitability long ago. There are too many traders--let alone quants with massive data mining capabilities--to not discover something this simple. I knew of a seasonal strategy that performed quite well through 2008 (it actually made money that year, going long-only). There was even a top-ranked newsletter that promoted it. Since then, it's performed terribly. That was the longest-running "stupid simple" system I'm aware of that beat the S&P 500...for quite a few decades. It's nice to think something really simple could work (and many gurus and trading coaches tell you this), but trading just isn't that easy. Maybe it was in the 70s and early 80s, but back then only a handful of people did any backesting with computers, and they were much more cumbersome and limited...so it kind of washed out unless you were Ed Seykota or something.
I found that generally true on my backtest results, especially strategies that depends on technical analysis like SMA, MACD, RSI.... Usually worked short term but going out a decade or more, their results tended to be mediocre. As I said many times, for us mere mortals making money trading is really hard.
Been there done that. I tested the following, from 1993 to 2019, ignoring commissions, bid/ask: 1. Sell at open today, buy at close yesterday 2. Sell at close today, buy at close yesterday 3. Sell at close today, buy at close a week ago 4. Sell at close today, buy at close a month ago 5. Sell at close today, buy at close a year ago 6. Buy in 1993, sell in 2019 6 > 5 > 4 > 3 > 2 > 1
#1 on your list is what the article was referring to. Just make sure you're testing this as a LONG-ONLY strategy. BUY at the close of today then close the position @ the open tomorrow. Basically "LONG" the overnight session & no position during regular trading hours.
Using yahoo's S&P 500 data (yahoo symbol ^GSPC), I ran a test with the logic for this strategy as: If (today is the third Friday of the month) or (today is the day before the third Friday of the month and the market is closed tomorrow), Then enter long at the close today. Exit at the close the following Friday or the next trading day if the market is closed the following Friday. The results were not good. For July 15, 1983 through July 26, 2019, some statistics on the simulated trade results in S&P 500 points without accounting for trading costs were: numValues 433 sum -90.5598780000002 prod -Inf min -163.75 max 96.209961 mean -0.209145214780601 sampleStdDev 24.3658056373813 median 0.19000299999999 medianAbsDev 8.40000800000001 geomean NaN skewness -0.981992731872224 excessKurtosis 7.2990019077746 >Thresh_0_Pct 51.27 The attached enter_long_optexp_one_week.xls has individual trade results.