The Forecasting Systems Letter
Jeffrey Mishlove

---

The figure below, from an Excel chart, shows the absolute, average price movement of S&P e-mini contract, during the 31 trading days from Sept 12 to October 24, 2002, for fiften minute bars between midnight and 9:30 pm, eastern time.  Because, this is a chart of "absolute" price movement, price direction is not indicated. But, I believe that this general, intraday pattern of price movement would apply throughout the year. 

During the evening, for various reasons, a trend can develop.  The trend may accelerate at about 3 a.m., eastern time, when the German DAX market opens.  Then, when the East Coast traders awaken, they will often place some orders about an hour or two before the opening of the trading day.  Of course, even more orders are place at the open itself.  Often these orders are based on the emerging trend.  So, by anticipating the trend as early as midnight, one stands to profit as the trend builds up to the market's open.

Here is a simple, midnight trading strategy that looks promising:  All trades are placed shortly after midnight (eastern time) and are closed at 9:30 a.m. (eastern time) when the market opens.  The trading signal is simply whether whether the market is up or down from the close of the previous regular trading session.

I have backtested this system, using the December S&P e-mini contract, between September 5, 2002, and October 24, 2002.  During that time, the system made 36 trades, of which 21 (or 58.33%) were profitable.  The average winning trade gained 7.48 points or $363.80 on the e-mini contract.  The average losing trade lost 6.27 points or $313.33 on the e-mini contract.   The total profit, trading one e-mini contract came to $3290, or $91.39 per trade.  These figures include $10 per trade for commission and slippage. 

Had the system been set up to trade one regular S&P contract, profit would have been $16,450 for the 36 trades -- or $456.94 per trade -- allowing $50 for commission and slippage.

Here is a simple morning trading strategy that also seems promising to me.  The trade signal is developed by subtracting the closing price of the 8:15 am, fifteen minute bar from the closing price of the 9:15 am, fifteen minute bar.  This was done on the S&P e-mini contract.  In other words, if the price has dropped during the hour prior to the open of the regular trading day, we go short.  Otherwise, if the price has risen, we go long.  In this system, we close our position at 10:15 am.

In backtesting this system, with Excel, from September 12 to October 24, 31 trades were placed, of which 21 (or 67.74%) were successful.  The average winning trade yielded 5.18 points, or $258.93 on the e-mini S&P contract.  The average losing trade lost 4.375 points, or $218.75 trading the e-mini contract.  The total profit from 31 trades was 65 points of profit.  If one were trading a single, e-mini contract, the total profit would be $2940, which includes $10 per trade for commission and slippage.  The profit per trade would be $94.84.

Had the system been set up to trade one regular S&P contract, profit would have been $16,450 for the 36 trades -- or $474.19 per trade -- allowing $50 for commission and slippage.

Note that these backtests did not include any provision for stops.  I believe that the system's profitability might  be improved through their use.  I would recommend the MAE/MFE system developed by John Sweeney for this purpose.  In a future edition of the Forecasting Systems Letter, I will discuss the use of this system further.

During the period in question, the market was quite volatile, with strong directional movements.  However, since the market ended where it began, there was really little overall trend.  Or, one might say there was a strong downtrend, followed by an even steeper uptrend.  The actual, daily candlesticks for the S&P contract are shown below in a chart created by OmniTrader.  One can clearly see the gaps in price movement between regular trading hours.

It is understandable that there would be both upward and downward bars in the directional chart shown below -- given that the market showed strong movements in both directions during the period in question.  The chart below, created in Excel, is almost identical to the Excel chart at the top of the above article.  The major difference is that the price change is no longer absolute.   And, obviously, this is a bar chart, rather than an area chart.  The average price shown, for each 15 minute bar, takes into account the direction of the price movement.  The same, 33 day time frame is used as depicted in the candlestick chart above.

I find it very interesting that there are eleven consecutive bars that show upward movement from 11:15 a.m. to 1:45 p.m.  While more data is obviously needed, this sampling suggests that the market tends to quietly drift upward during the the time that East Coast traders take their long, power lunches.  The relative quietness of the market at this time is well-known to traders.  And the following Excel chart, clearly shows the diminishment of trading volume during the lunch hour period.

These patterns suggests the possibility for small, scalping trades, within a narrow, slightly upward sloping channel during the lunch hour.  In a future Forecasting Systems Letter I will provide data regarding tests of this hypothesis.

Finally, allow me to show you a trading system that takes advantage of the very strong correlation that exists between the German DAX index and the S&P500 futures contract.  The chart shown below is of the same type as I have used often before in this Letter to demonstrate the effectiveness of various leading indicators.  However, we cannot regard the German DAX as a true leading indicator, because its trading overlaps with the S&P contract.  Trading on the DAX starts at midnight (Pacific time) and ends at 8 am (Pacific time). 

In order to fully understand the meaning of the above chart, it is useful to see what perfect correlation would look like.  And, of course, the S&P500 contract is perfectly correlated with itself.  This is shown in the chart below:

The chart just above shows that you would have earned $4,113,800 since January 1, 1997, had you made daily trades of a single S&P contract with perfect accuracy.  In this instance, the trading signal -- shown in yellow -- is identical to the trading target: whether or not the conract price will move up or down following the next day's close.  Of course, it is impossible to use the perfect signal (unless one has precognitive access to the future, which I regard as problematic but not impossible), since the signal is only available at the time the market closes. 

The chart that uses the German DAX to provide a trading signal shows that it would have earned $1,961,775 trading the S&P futures contract.  In effect, the German DAX signal is 47.69% as good as a perfect signal.  This is very interesting.  A good neural network signal might result in profits around 20% as good as a perfect signal.  That would be highly profitable.  The German DAX signal is twice as good as that!  There is simply one catch:  By the time the German DAX signal is available, the S&P contract has only five hours left to trade.  Undoubtedly, much of the correlation shown above is already incorporated into the S&P contract price by that time.

To see if there was any value left in the German DAX signal at the time of its close (8 am, Pacific Time), I backtested a trading strategy that used the DAX signal to place a trade at that time, closing the trade at the S&P close.  It actually worked quite well.

The system placed a total of 36 trades between September 6 and October 25, 2002.  22 of those (or 61.11%) were profitable.  The average winning trade gained 10 points, or $2500 trading an S&P contract.  The average losing trade lost 9.54 points, or $2383.93 trading an S&P contract.  The total profit for all 36 trades was $19,825 after deducting $50 per trade to cover commission and slippage.

How does that compare with a perfect signal?  The chart below, shows "perfect profit" during the same trading period. 

As you can see, perfect trading would have resulted in a profit of $137,125.  If we deduct $1,800 to allow for commissions and slippage, that leaves $135,325.  So the very respectable, backtesting profit of $19,825 is 14.65% of "perfect."  In general, this also suggests that about 30% of the value of the German DAX correlation signal still remains to be harvested after the German DAX market has closed. 

Incidentally, when conducting the German DAX backtest, I also compared the German DAX signal to the signal created by the direction of the S&P contract itself at 8:00 am, Pacific time.  During this particular testing period, the German DAX signal outperformed the S&P contract signal by more than 20%.

After my recent trip to Albuquerque, it has become more than evident to me that it will not be possible to continue publication of this letter on a daily basis.  My travel schedule will not permit it.  My other competing business interests will not permit it.  My other professional interests will not permit it.  But, most importantly, the care and diligence required to prepare worthwhile articles and analyses is not consistent with a daily letter.  Therefore it is reasonable to expect that -- from this time forth -- publication will be more intermittent.  I expect it will be closer to once a week than once a day.

Also, I have now scheduled travel to Istanbul, Turkey, from November 9 through Nov 20.  I expect that travel will interfere with posts to this website.

BioComp Profit Neural Network S&P 500 Futures Contract "MegaSystem" Forecast:
Wednesday's Close: Down from Tuesday's close
Thursday's Close:  not yet available

Nirvana OmniTrader Composite Technical Forecast for the S&P 500 Futures Contract:
Short-term, aggresssive strategy: Down.

NeuroShell Volatility Breakout System, Daily, for S&P Futures Contract: 
Buy Stop Order for Wednesday at 894.90.

KWIK*POP Daily Trading Signal (As Confirmed by Hourly Chart)
Out of Market since Thursday, October 24

View the Results of Previous Forecasts
Forecasting Systems Letter Archive
Return to ForecastingSystems.com