Possible off-by-one error in historical MTI records!

[catpees]

Hello everyone,

I am quite sure that there is an off-by-one problem in the historical MTI data, whereby the MTI Signal and the MTI Alert Level are appearing on the previous day's record. Take a look at what I mean with the example below.

Let's take a look at some historical data where the signal changes a couple of times:

Code:

Date         TSX Close    MTI Signal   MTI Alert Level
2015-11-30   13469.83     down         13595.816
2015-12-01   13636.06     up           13440.882
2015-12-02   13463.82     up           13437.203
2015-12-03   13324.67     down         13532.314

And now, let's look at the emails for those same trading days:
Orders for the 2015-11-30 (received Sun, Nov 29, 2015 at 8:46 PM)
Our Market Timing Indicator is DOWN. The alert level is set to 13617.86
Orders for the 2015-12-01 (received Mon, Nov 30, 2015 at 11:51 PM)
Our Market Timing Indicator is DOWN. The alert level is set to 13595.82
Orders for the 2015-12-02 (received Tue, Dec 1, 2015 at 9:48 PM)
Our Market Timing Indicator is UP. The alert level is set to 13440.88
Orders for the 2015-12-03 (received Wed, Dec 2, 2015 at 9:18 PM)
Our Market Timing Indicator is UP. The alert level is set to 13437.2
Orders for the 2015-12-04 (received Thu, Dec 3, 2015 at 11:06 PM)
Our Market Timing Indicator is DOWN. The alert level is set to 13532.31

Notice how the historical record for 2015-12-01 shows the signal as being UP, but the email showed the signal as being DOWN for that day. Also notice how the historical record for 2015-12-01 shows an alert level of 13440.882, but according to the live signals (i.e. email and web page), that was the alert for 2015-12-02.

So anyone using the historical MTI data should take a closer look to make sure it reflects the emails and/or webpage updates. The quick fix in Excel is to shift all of the MTI Signal/Alert values down by one row.

If I am off my rocker, please chime in...

Cheers!

[catpees]

When using the historical data for back-testing, this off-by-one issue has us moving to cash one day earlier than the actual DOWN signal, and moving to equities one day earlier than the UP signal -- this makes back-testing show very favourable results that don't reflect reality.

I hope this helps.

[Super Stock Picker]

Hello Catpees,

First, welcome to this forum!

You have noticed a very important point about the MTI signals, this is indeed key to use it properly in order not to skew your backtest results. A one-day shift makes a huge difference as the indicator usually changes after a big move in the markets.

So, here is the key point:
The MTI changes take place at the close of the day.

If I take the value from your excerpt, that would give:
- for 2015-11-30, the MTI is down and the alert value is set to 13595.816. This alert value is the threshold that will be in place the next day, ie 2015-12-01.
- on 2015-12-01, the TSX closed at 13636.06 ie above the alert value. So, the new MTI vaue is UP. At the close of that day, the portfolios are buying all the stocks back.
- a new alert value is set for the next day at 13440.882

I hope that explains how we report our historical values.

[catpees]

Hi SSP,

Thanks for the explanation; it helps. I guess the thing I found difficult to grasp is why the alert level for a given day would be for the following day's close. Knowing now that the historical data records the MTI Signal and Alert Level the same day it was generated makes sense. I don't think it was clear in the original MTI description, especially since the very first Excel file reports the MTI Signal/Alert on the day it applies to rather than the day it was generated.

Thanks for the clarification!

Cheers!

[catpees]

Hi SSP,

I just wanted to mention that the historical MTI data has two entries for June 22, 2012. After looking at actual market data, I can confirm that the first instance is actually for June 21, 2012 (i.e. the one which lists a closing price of 11408.32).

Cheers

[Super Stock Picker]

Hello Catpees,

Thanks a lot for catching that one!

It has been fixed by today's release.