Desk Notes
The Indicator I Deleted and the One I Kept
Two panes agreeing feels like confirmation. If both panes are built from the same closing prices, it is one piece of evidence counted twice.
I deleted the stochastic oscillator. Moving average convergence divergence went a few months later. Average true range I kept. It tells me nothing about direction at all. That is the answer. The reasoning took one afternoon and a column comparison. That part is worth more than the answer.
What each pane is actually made of
My charts used to carry three momentum indicators stacked down the right hand side. I had a reason for each. This one catches turns early. This one smooths the noise. This one shows the trend behind the momentum. Three reasons, each arrived at separately and each perfectly good on its own, which is how you end up with a screen that nobody designed and that nobody would have designed if they had sat down to do it deliberately.
| Indicator | What it consumes |
|---|---|
| Relative strength index | Closing prices over a lookback window |
| Stochastic oscillator | Closing prices over a lookback window |
| Moving average convergence divergence | Moving averages of closing prices |
| Average true range | The full range of each bar, including gaps |
| Volume | Shares traded |
Written out like that, the problem is visible before you run anything. The top three columns are all the same input wearing different clothes. The relative strength index normalizes with average gains against average losses. The stochastic normalizes by where the close sits inside the recent high and low. Different arithmetic, one source.
So I put them in columns
A few hundred daily bars, eight instruments I actually traded, both series exported side by side, and a column that did nothing except count the days on which the two of them disagreed about whether the instrument was overbought, oversold, or somewhere in the middle.
There was no such day. I could not find one where the relative strength index sat in its upper zone while the stochastic sat in its lower one, and the moments they did differ were moments when neither was saying anything I would have acted on. Mostly a bar of timing.
I should be honest about the quality of that test. Eight instruments I chose because they were already on my screen is not a sample, the lookback settings were whatever I happened to be using, and I never checked a second period. It was enough for my purposes because the result was not close. Had it been close, I would have had to do the work properly.
The uncomfortable part is what redundancy does downstream. Two panes agreeing feels like confirmation. Confirmation raises confidence. Confidence raises position size. And none of it has any foundation. The second opinion came from the same source as the first. I had built a machine for manufacturing agreement with myself, and then I had wired it directly to my sizing, which is a considerably worse thing to do than simply owning too many indicators.
The question that replaced the reasons
What does this show me that something already on the screen does not?
Price and structure answer it. Where the level is. Whether it held. What the bar did when it got there.
Volume answers it, because participation is separate information from direction. A breakout on thin volume and the same breakout on heavy volume are two different events at an identical price, which is the argument made properly in volume analysis.
Average true range answers it. It is the one I would keep if I could keep only one. It says nothing about which way anything is going. That is exactly why it earns its pane. It tells me how far this instrument normally travels in a day, which tells me where a stop is far enough away to survive ordinary noise, which tells me what size the position can be for a fixed risk budget. Three decisions out of one number. Nothing else on the chart duplicates any of them. The calculation, and how it interacts with band based tools, is in Bollinger Bands, ATR and Keltner Channels.
Moving average convergence divergence survived longer, because it carries trend information the oscillators do not. It came off anyway. I had noticed I was reading it as a slower version of the same momentum story, and that I had never once acted on it against what the others said. The arithmetic behind all three, and the honest account of what divergence does and does not predict, is in RSI, MACD and stochastics.
Run it on your own screen if you want. Export two series for the same instrument and period. Count the days they disagree. Twenty minutes. After that the argument is settled for good. You stop having it every time you add a pane.
My charts are emptier now and my decisions are slower, in the specific sense that there is less on the screen telling me I am right. The premise and the limits of the whole approach are set out in technical analysis for beginners, and the section on what charts cannot tell you is the one I reread most.
Frequently asked questions
How do I know if two indicators are redundant?
Export both series for the same instrument and period and count how often they give opposite readings at the same moment. If one almost never says anything the other has failed to say, you are looking at two presentations of a single input rather than two independent pieces of evidence.
Why do RSI and stochastics behave so similarly?
Both take recent closing prices over a lookback window and express where the current close sits within that range, so they are transformations of the same underlying data. The formulas differ and the smoothing differs, which changes the timing of the signals more than it changes their direction.
What does average true range actually tell you?
Average true range measures how far an instrument typically travels in a period, including gaps, so it describes volatility rather than direction. It is most useful for deciding how far away a stop belongs and how large a position can be for a fixed risk budget.