Levels are everywhere in stores. There are thousands of ways to find, define, use, and abuse layers. Many levels simply have no advantage.
In this blog post, I’d like to share some of the work we’ve done recently for our MarketLife members, and I’ll give you some concrete ideas on how to use levels in your trading in an upcoming post.
My work with levels
I had a number of levels that I had developed and used in my intraday trading in the early 2000s that were based on volatility, previous bar relationships, and a longer-term trend perspective. When I started doing core statistical work, I found that it was very difficult, if not impossible, to verify that most of the levels had an edge over random expectations. (Though this research shows some promise!)
I also became more aware of cognitive biases and saw that any level drawn on a chart at any one time would matter to a trader. Remember, this was almost 20 years ago, but the situation hasn’t changed: most of the people who talk about levels do so after the fact, and say things like, “Oh look again, this market is on a Fibonacci -Level stopped. Isn’t math great? “But of course this thinking is not strict and nothing is ever really critically examined.
I found that I could trade very well without the levels and gradually eliminated all but the most obvious levels from my trading process without negatively affecting my results. Momentum and market structure were far more important. However, I’ve always wondered if maybe I wasn’t missing a thing … maybe I threw the baby out with the bath water?
In the past few years, I’ve done more work developing some new methods of evaluating levels. (This blog post was an interesting part of that process, using human discretion as an input for statistical analysis.) When I revisited my old levels, I found that they worked very well … maybe amazingly well. It’s not an exaggeration to say that most of the levels are statistical junk, but they do have a real benefit.
Meet the PowerLevels
The problem with these levels is that there was a wide variety of them that could be in the game any day – up to 20+ and these grouped within about +/- 3 ATR of the previous bar. In practice, I never used all of them, but I had a complex mix of discretion and rules to tell me which levels are likely to be in play in a day.
In the last few weeks (from October 2020) I have tried to fully automate the level selection process as I have free time thanks to the ongoing restrictions of the coronavirus (!). I won’t go through all of the details (nor will we reveal the insides of the layers in detail), but the general idea I worked with was that the algae needed to mirror my thinking process as much as possible.
We’re still working on it a bit, but we’re mostly there. We have been using these levels internally for several weeks and are very happy with what they have done for our trading. We used them to set stops, refine goals, and infer general direction dependencies for the day.
We currently use two sets of levels: Every day we generate intraday levels based on the previous day’s closed bar. We publish these levels as part of ours PreMarket Edge report. (Which is pretty close to the best value in the business, by the way. For $ 19 per month (less than $ 1 per trading day) you can get these levels, a quick pre-market video of my insights and my schedule for the upcoming session, a carefully crafted list from Premarket Movers to watch out for and more.)
The second levels are more important for the daily structure. These levels are generated before the start of the week and apply for the entire week.
A quick test
You should be wondering if these levels are better than random levels. (This is always the question that must be asked of every technical tool.) The answer is a resounding yes. Here is an example: Look at the intraday values (shown as dots) in the table above. Remember that these are generated at the end of the previous bar, so they are fully known before the day starts.
An occasional look at the table above shows that there seem to be many bars where “almost” the exact high or low of the bar was “a level”. At first glance this looks pretty cool, but wait – we’ve drawn a lot of layers in the table. Maybe – actually probably– It’s just a misfire of our pattern recognition machinery, and it looks better than it is. That should be your standard thinking.
One way to test is to compare the actual PowerLevels with randomly generated levels. This requires a little thought, but we decided to let the PowerLevels define the area for the day and then distribute randomly distributed levels over this area. (One of the subtle, but very powerful, aspects of the real world levels is that they react very quickly to changes in volatility. You can see from the table above that the levels are sometimes very wide and sometimes narrow.) If we run the test without this Condition where the random levels are simply scattered over an area, the random levels are destroyed by the real levels. If you give this information to the random levels, that’s at least a fair game.
However, the random levels are still losing, which shows that the PowerLevels actually define the next day’s highs and lows more often than expected. Looking at a representative sample of active markets, we find that the closest PowerLevel is within an average of +/- 0.23% of the final high / low for that session. In the case of random values, this distance is almost three times as large with +/- 0.63%.
This is just an illustration, but given the other work we did and how long these layers were part of my toolkit, they strongly suggest an advantage.
Of course, using a level means more than just knowing where it is. And that’s what we’ll cover in the next blog post! Stay tuned…