# Category Archives: Binary

This was one of the things that just bugged me to the point where I needed to actually sit down and figure it out. Here is the complete picture:

The data is pretty straight forward, recording 2 extremes (high/low) within each “hour” and storing them into a 10 minute bin. The question is: Why does the area where the extremes are vary so much from hour to hour? Personally, I had a hard time seeing the relationship and working it out in my head until I had all of the sets complete. Can you see a pattern?

When I think about extremes, I think about what I call the leading extreme, which is more or less the same as the “trend direction”. Take the following snapshot, where i’ve deleted the time stamps so that we can make them up.

Let’s assume that we are looking at a typical case where the yellow lines are representing xx:00; That is, 33% of the time an extreme is located in the last 10 minutes of the hour. The yellow circle is representing that the extreme is indeed in the last 10 minutes. Great!  Now if you imagine the vertical yellow lines shifting over 10 minutes, now starting at xx:10, we can see that price dips down, and the high extreme is still located at that peak, which is some time between xx:50 and xx:00. So why is the probability 33% in some cases and as low as 12% in others?

Because of the trailing extreme. Let’s take a look at a side by side:

Again, assume the times of the first extremes to be xx:00-xx:10 and xx:50-xx:00 as shown by the yellow circles. Now if we shift the time forward by 10 minutes, the extremes are now xx:10-xx:20 and xx:50-xx:00. The “leading” extreme time has not changed, but the first extreme (the low) has now been shifted 10 minutes forward.

This means that when price is like above, in a healthy/strong trend, or occurring in an area where price moves in one direction only for a period of over 10 minutes, it creates a varying extreme on one end, even if the other end (the high in this case) remains the same.

The big question now: Does it matter? Maybe.
I mean it’s not completely insignificant. The biggest probabilities within each set are not that small, at least twice as big as the ones that are least likely. If you take the two largest probabilities in each set (which are always next to each other), the number ends up being close to or over 50%. That is, no matter what time you use as your start time, a particular 20 minute period will contain an extreme 50% of the time. That is, 33% of time will contain 50% of extremes. That’s pretty cool!

With this understanding, the relationship is a bit clearer now. Given how the trailing extreme functions, The higher probability numbers are occurring where we expect them to be: 10 minutes trailing.. for the most part.

There’s something that sticks out very clearly, and that’s the pause that happens from the xx:30 start to the xx:40 start. Remember that the binned number is the max. This means that for both those periods, the minutes of xx:20-xx:29 contain an extreme more often than any other 10 minute chunk. Therefore, the stats seem to favor the xx:20-xx:29 period as the best period to be making extremes of all time frames by a slight margin.

# Hour Compositions Pt 5: Wave types by time

This one is pretty self explanatory. I like the idea behind it, but there is definitely not enough data for this one to be too meaningful. I’ll have to come back to this in a couple months (or even years!) The difference between A vs B waves to be more likely by time actually makes a big difference, so I’m interested to see how it turns out.

My future ideas are mostly grounded in studies of A waves vs B waves. This can be taken in multiple ways, either in studying the difference between the two to know which one is most likely to occur, or studying them individually to find out a way to best trade each type.

# Hour Compositions Pt 4: Winners by time elapsed

Just waiting and collecting more data now.

Here I took a look at the trades that had 1 or more reversal points (aka, BCDE waves). Time elapsed is measuring the time from the ‘hour’ start until the point where the wave can officially be deemed a “at least a B wave”. This type of study isn’t reallly for trading purposes (although it certainly can be) but it’s one of those tidbits that jive well with the story of trending and reversals. We can see that the best win rates are among those with about half an hour elapsed or more.

Win rates in this case correspond to the probability that price follows the original trend.

If the green line is the final outcome, than the signal is a “win”. If instead it is the red line that occurs, it is a “loss”.  Why does more time elapsing (to a certain extent mind you, because 45+ win rates decrease slightly) lead to a higher probability for price to close in the original direction?

I think this is in part due to the nature of taking 1 hour snap shots. That is, this data does not take into account any past bars or data. As a result, the “trend” that we witness originally may be a new trend, in the middle of a trend from the last hour, or the end of a trend. However, if we make the assumption that trends… well.. trend, then more time elapsed in a trend equates to more distance. I think what this data may be suggesting is that while price moves up and down, if a trend survives 30 minutes (plus whatever from the previous hour), the next move (in the opposite direction) will not be strong enough to over take it. In other words, this could be again helping the trend theory.

# Hour Compositions Pt 3: ABC waves in the 1 hour context

I tried to use my old old ABC waves to confirm that these kinds of studies are indeed reflexive in any time frame.

Same as with the old model, I added a simple time filter to normalize the data a bit more while maintaining most of it’s integrity. Most important to me is how ABC waves still make up the majority of waves, and I can use this as a method for tuned entry in some scenarios, especially when one can make money betting that price will NOT be at a certain price or range of prices. The “trend” direction here actually has a slight edge (A+C+E= 55%) which is nice. I generally don’t count anything under 60% as an edge, but in these very micro frames, it counts for something. I still need more data to be comfortable with it, but it’s a start.

Using this trend following idea, I’m trying to look into dynamic “safe” levels. When dealing with hourly data (or intra hourly really), I think time filters are mandatory.

I have a bit more work to do in trying to optimize this, but early results do seem to suggest that active hours are more even in terms of trending moves while off hours favor range bound movements. Seems correct.

# Hour Compositions Pt 2: Strength over time

I pulled up the statistic that I had done a while ago about the effect of the current bars movement on the net major move of the day. As it stood in that version, a movement of 25 pips in a single direction gave about a 60% edge to the major move of the day being in that direction.

Given my recent interest momentum, toying with the idea of SB work and using small frames as my data, I wanted to research this idea a bit more thoroughly. If you measure the distance from a certain close to the current open, how likely are you to be able to predict the close of the higher frame relative to the open? That is, if I’m looking to determine the close of the hour relative to the open (is C-O > 0 or is C-O < 0 ?), is there a point where I can know for sure if the answer is one way or the other? Initially my thought process was that if I know where the 58th bar closes relative to the open of the 1st, I can probably know that the close of the 59th bar relative to the open of the 1st bar will be the same. Let’s run the data:

Most interesting to me was the strength that one gets from even a single bars close. By the time you wait 8 or 10 bars, or 1/6th of the way through, you have what is in my opinion seemingly substantial evidence for momentum. I also took a look at the few cases where even knowing the second to last bar cannot tell you where the next bar will close. These are not cases where the last bar makes some crazy move, rather those bars tend to sit around the open price to begin with, letting just a few pips in either direction make the difference.

This made me think: If a bar goes from 0 to +5, it makes sense that it would be likely to finish somewhere in the positive region; after all it is now “cheating” by having a +5 buffer. Okay. What if I tested this dynamic looking at cases where price moves to +5, then back down to 0 or in the negative space? At this point, is price still likely to finish in the positive space?

Not quite. At this point, the edge becomes more like a traditional coin flip. It is good to know that the stats are pretty consistent across the board, even with the relatively small sample size (2 months). On the second test, the probability actually does decrease over time, from 50% likely down to 46% likely. This is perhaps suggesting that when price waits a long time before breaking a price line, it’s a signal of reversal. However, 46-47 is not strong enough to say that this is a reliable signal

While expected (since we know how good the market is at being 50-50 with these things), it’s actually not a bad thing to know in the end. The stats are in some way confirming support and resistance. As long as price stays above the open price, price is likely to end above that price. So if one gets the chance to get in on price when it is on a pullback then at best, there’s generally a 10%+ edge to be had. At worst, it’s a coin flip. Therefore, as long as one is trading with at least a 1:1 ratio, the coin flip losses are perfectly fine, and the edges shine.