Learn the terms · Longer read
Backtesting: replaying the rules over the past and counting the misses
Replaying the outlook's readings over a stock's past prices and counting, for every day they had a view, whether the price was higher, lower or about the same 20 trading days later. A 10-day count is computed too; the page shows the 20-day one. A record of the past with all its misses; not a prediction.
What it measures
For each stock you hold or watch, Trendentum takes about four years of daily prices and, from the first day with a year of history behind it, runs the same readings the live outlook uses, with the same fixed settings, over the same 300-day window the live refresh fetches - so the test cannot quietly use a different engine from the one that produces today's read. Every day a reading has a view is a row; every Positive or Negative outlook is a row. Each is then scored by where the close sat 20 trading days later (a 10-day count is computed too; the page shows 20): moved its way, moved the other way, or stayed within 1% - "flat", so a coin-flip reading cannot look half right on noise.
Mixed days are not scored, because "no clear view" cannot be wrong. The insiders and short sellers reading is left out and the panel says why: there is no history of filings or short reports on file to replay. A reading holds its view for weeks, so neighbouring days overlap; the "clear of luck" test counts about one independent test per 20 days, not one per day. What comes out is a table of counts per reading and per outlook, and a list of the worst misses - the days a Positive or Negative outlook was followed by the largest move against it, with the date, the readings behind it and the move.
Why the misses come first
Most published backtests show a rising line and a percentage. Trendentum's shows neither, by design. A compounded return assumes you acted on every call at the close with no costs, no slippage and no hesitation, which nobody does, and it turns a description of the past into an implied promise about the future. The payload has no such number in it - not hidden, absent - and a test fails the build if one appears.
Counts are honest in a way rates are not. A line like "the Negative outlook on this stock was followed by a higher price 61 times out of 91" tells you something specific about the readings on that stock; "67% win rate" invites you to imagine a portfolio. The worst misses come first because they are the part a sales page leaves out and the part that tells you what the rules are like to live with.
What it is not
A backtest is not evidence that a rule will work next month. Prices are not stationary; the conditions in which a rule did well may not be the ones ahead. Three years is one slice of history, and a stock listed for eighteen months has less. The counts also assume the readings' settings were fixed in advance, as they are here: the readings use fixed settings, and your strategy settings change only the chart figures, never the outlook or the backtest. Tuning settings until the past looks good is the classic way to be misled, and this design leaves nothing to tune.
The private backtester exists so you can see how the rules have behaved on what you own before you weigh what they say today. It is never shown on a share card, in an email or on any public page, and Trendentum publishes no track record. A record that includes the misses is useful to its owner and misleading as an advertisement.
An example
An illustration of how to read a count. Suppose the panel shows this line for one holding: the Negative outlook was followed by a higher price 61 times out of 91 over 20 trading days. Read one way, the readings were wrong two times in three on that stock. Read another, the stock spent those years rising through every reading of weakness, which is also information. Either way the number is a count of what happened, not a rate dressed as a promise, and it is the kind of line the miss-first design exists to surface.
Where you’ll see it on Trendentum
- "How the readings have done on" and the stock code, on every stock page
- "How the rules have done on your stocks" on the dashboard - counts across your book, the ten worst misses
- The methodology page, on what the test never produces
Every term on this page is used the same way across the app, and the exact thresholds behind each rule are on the methodology page. Trendentum explains every call and never tells you what to do with it.