At the bottom of the Distributions tab there is a monthly returns heatmap, a grid of twelve month cells and a total. On the capture I am working from, eight of the twelve are blank. April reads -0.4%, May reads -20.5%, June reads -8.1%, July reads -16.8%, and the total reads -45.8%.
A heatmap is a seductive object. Colour makes any grid look like a finding, and the human eye is extremely willing to see a pattern in four coloured squares. So it is worth being precise about what four cells can and cannot support, because the answer is narrower than people assume and there is something better to read in the meantime.
The one reading four cells honestly supports
Four months, all negative, in sequence. That is the finding, and it is a real one.
How impressed should you be by four negatives in a row? If months were coin flips with an even chance of finishing positive, four negatives in a row happens about one time in sixteen, which is a bit over six percent. That is suggestive. It is not conclusive, and anyone who tells you a four-in-a-row streak settles the question is doing arithmetic on a sample of four.
The magnitudes carry more information than the signs do. A month at -20.5% and a month at -16.8% are not marginal misses. Two of the four months lost a fifth and a sixth of the book respectively, and that is a statement about position sizing and stop discipline that does not need a larger sample to be worth acting on. A run of small negatives is a strategy having a quiet spell. A run containing two double-digit negatives is a risk question.
The three things it cannot tell you yet
The first is anything about calendar seasonality. To say "May is a bad month for me" you need several instances of May, and a twelve-cell grid covering one year contains exactly one of each month. Four filled cells is four observations of a strategy in sequence, not four observations of a calendar. This is the mistake that a heatmap layout actively encourages, because arranging returns by month-of-year visually implies that month-of-year is the variable, and with a single row it cannot be.
The second is anything about the within-month path. A cell reading -0.4% for April is consistent with a month in which nothing happened and equally consistent with a month that fell twelve percent, recovered, and closed almost flat. Those are very different months for anyone who had to sit through them and for anyone whose stops were being tested. Monthly aggregation destroys exactly the information that risk questions need.
The third is the difference between blank and zero. Eight cells on this grid are empty, and empty means no data, not a flat month. Skim a coloured grid quickly and a blank cell and a zero cell look nearly identical. It matters, because August is blank here while the account equity chart on the overview tab of the same page runs on into late August and the month-to-date tile reads 0.00%. Blank cells and zero cells sitting side by side is a good reason to slow down and check which is which before drawing any line through the year.
The total row and how it was aggregated
Here is a piece of arithmetic worth doing yourself, because it teaches something general.
The four monthly figures are -0.4, -20.5, -8.1 and -16.8. Add them up and you get exactly -45.8, which is the total shown. Now compound them instead, which is what actually happens to money: multiply 0.996 by 0.795 by 0.919 by 0.832, and you get 0.605, or about -39.5%.
Two aggregation conventions, six percentage points apart on the same four cells. Neither is a mistake. Adding period returns and chaining them are both standard, they answer slightly different questions, and they diverge more the larger the individual returns are. On four small months the difference would be a rounding error. On a run containing a -20.5% month it is six points.
There is a third number in the neighbourhood, because the risk block on the overview tab of this same page shows a year-to-date figure of -38.59%. So you have -45.8% on the heatmap total, about -39.5% if you chain the same cells yourself, and -38.59% on the calendar tile. Three figures for "the year".
I am not going to tell you which is correct, because from the outside I cannot verify what series each one runs on and guessing would be worse than useless. What I can give you is the list of general reasons two honest figures like these diverge, which is short and covers almost every case you will meet:
- Additive versus geometric aggregation, as above.
- Different underlying series, in particular a sum of trade-level results against a return computed on an account equity curve.
- Different populations, since a page like this has toggles for real against paper, all accounts against one account, and one source against another.
- Different windows, where one figure covers all available history and another covers the selected date range.
- Cash flows, since a raw balance series moves when you deposit or withdraw, while a time-weighted return series is built to ignore that. This is a big one and it is easy to check.
The export test that settles it in fifteen minutes
You do not have to accept any of this on faith. The page has an export control in the header, next to the share button.
Set the account scope, the source filter and the date range explicitly first, so you know exactly what you are pulling rather than inheriting whatever the session had selected. Export. Then in a spreadsheet, build the month-by-month returns yourself from the equity series, and separately sum the trade-level results by month. Compare both against the grid.
Three outcomes. If your equity-derived months match the grid, the heatmap runs on the equity curve and you now know that. If your trade-sum months match instead, it runs on trade results, which behaves differently the moment you deposit or withdraw. If neither matches, look at your deposits and withdrawals first, because that is the most common single cause, and then at whether the date boundaries are being cut in your timezone or somebody else's.
Whatever you find, write it in a note next to the number. Not because anybody is going to audit you, but because in three months you will look at this grid again and you will not remember, and the fifteen minutes will have to be spent twice.
What to read while the grid fills up
The rest of this tab is a better use of your time right now, for one reason: sample size.
The trade distribution panel on this same tab reads n equals 87 on this capture, with a mean of -0.53 and a median of 0.00. Eighty-seven observations against four. The breakdowns on the overview tab split the same trades by direction and venue, and on this capture that split is genuinely informative: 73 long trades against 14 short, and three venues with very different hit rates behind them. Those are readings you can act on now.
Set a threshold for when the heatmap becomes worth reading and then leave it alone until then. Twenty-four cells, two full years, is where month-of-year comparisons start to be worth anything, and even then you have two observations per month. Until you get there, use the grid for one thing only: checking that each month ties to your account statements. That is bookkeeping, it is genuinely useful, and it works from the very first cell.