Seasonality

One of the Market Radar screens in RingMaster. It shows each covered name's own calendar record, month by month. Open Market Radar

At a glance

• The grid. One row per covered name. A covered name is one of the roughly thousand companies this desk works from each night. Twelve columns, January to December.
• What a cell measures. The change in the share price from the last trading day of the month before to the last trading day of that month. It is the share price on its own. Dividends are not counted in it.
• How far back it goes. As far as the price history we hold for that name. The screen prints the first and last year behind the figures.
• Average change. Add up what that month did in every year, then divide by how many of that month there were. One enormous month can drag it a long way.
• Middle change. Line all of those months up from worst to best and take the one in the middle. One enormous month cannot move it.
• Months up. How many of those months finished higher. It is written as a fraction, such as fourteen of nineteen.
• The count in the cell. How many months the figure rests on. It is usually a small number, and it is the first thing to read.
• Faint cells. A cell resting on fewer than ten months is drawn faint. The figure is still there. The faintness is the sample size warning you.
• Months left out. Some months are left out of the maths. The screen prints how many.
• It says nothing about next month. This is what already happened. Nothing here says what any coming month will do. Key difference. Average change moves with one big month. Middle change does not. When the two disagree sharply, one or two events are doing the work.

Read the counts before the figures

Plenty of people sell calendars promising the exact days to buy and sell. A grid of green cells looks like a discovery. Usually it is not one. The reason is arithmetic.

• Small samples. Nineteen Augusts is nineteen observations. Each one either finished higher or it did not, like a coin landing heads or tails. Thirteen heads out of nineteen tosses comes up by chance roughly one time in six. So thirteen higher Augusts out of nineteen proves very little.
• Many chances. Twelve months across a thousand names is twelve thousand cells. With that many chances, striking patterns turn up from nothing at all. They turn up in every dataset, every time. Finding one is not evidence of anything. So the screen is built so the sample cannot be hidden.

Reading a row

Three views show the same months: Average change, Middle change and Months up. Switch between them at the top of the screen. Click a row and the rest of the desk follows that company. Screens set to follow a name will switch to it. If the desk is already following a name, its row sits at the top of the grid.

Average change

The mean of that month's changes, in percent, across every year of the name's own record. One enormous month can drag it a long way. The view the grid opens on.

Middle change

The median of that month's changes. Line all of those months up from worst to best and take the one in the middle. One enormous month cannot move it.

Months up

How many of those months finished higher, written as a fraction such as fourteen of nineteen. The sample size is never hidden inside a bare percentage.

Best and worst month

Link a name and its best and worst month print above the grid, by average change. Each prints as a historical frequency. The month, how many of that same month finished higher, out of how many years, and the average change. Both are stated in the past tense and say nothing about the coming month. A month needs at least ten years behind it to hold either title, the same floor that draws a grid cell faint. A name with only one month clearing that floor shows a best month and no worst month, rather than the same month twice.

Two limits of the data

Survivors only. The price history we hold is of companies that are still listed. A company that went to zero in 2015 is not in it. So the figures in this grid are flattering. They are the record of the companies that made it, with the failures missing. The longer a name's record, the more of that missing history sits behind it. Calendar months, not rolling windows. The question is about calendar months, so the arithmetic runs on the calendar. The current month is left out until it closes. The line above the grid says which month the record runs through, and that is always the last one that has fully closed.

Months left out of the maths

Some months are left out of the figures. The screen calls those months set aside. Each cell says how many of its own months were set aside. The total across the whole grid sits on the screen. A set aside month is a statement about the records, not about the company.

What is deliberately not here

• No month is ranked or singled out. No ranking of names by their calendar record. No highlighting of a cell.
• No all-names line. An average of a thousand names' calendar records is a series we would have to build ourselves. It would read like an index and is not one. It also carries the survivor problem above at full strength. Each row is one name's own record, and the screen refuses to blend them.
• No significance star. A test statistic on twelve thousand cells invites the exact error this screen is designed against.
• No forecast, in any wording. The copy is past tense throughout.

Real world use

1. Checking a seasonal story you read elsewhere. Someone says a name is always strong in November. Look at how many Novembers that is.
2. Seeing how thin this kind of evidence usually is. The counts teach more here than the figures do.
3. Knowing a name's own history. What a company did in past Decembers is worth knowing. It is a fact about the past.

Frequently asked questions

What does a strong month in the grid tell me about the next one? Nothing on its own. What a name did in past Novembers is not what it will do in the next one. Why do the average and the middle disagree so often? One very large month moves an average and does not move a median. A sharp disagreement usually means the pattern rests on one or two events. What does a set aside month mean about the company? Nothing. It is a statement about the records, not about the company.

Related panes

• Rank Not Return, on what a ranking is and is not. It is the same discipline applied to Ohey's own numbers.
• Volatility History, a name's own record of implied volatility.