We wrote down what we expected to find using data up to December 2025. Then we measured 2026 with the real product. Here is what we got right, what we got wrong, and a trap we nearly fell into ourselves.
Imagine you buy something that promises to move three times as much as the stock market. Sounds clear. The problem is how often it does that multiplication.
It doesn't multiply by three what happens over the month. It multiplies by three what happens each day, and the next day it starts from zero wherever it left off. It's in the prospectus of every one of them, in those words or similar ones.
It looks like a detail. Watch what it does:
| Monday | Tuesday | In the end | |
|---|---|---|---|
| The market | −10 % | +11,1 % | 0,0 % |
| Your «3X» | −30 % | +33,3 % | −6,7 % |
The market is back exactly where it was. You've lost 6.7 %. And nobody did anything wrong: it's the maths of going down and coming back up. Falling 10 % and then rising 10 % doesn't leave you where you started, it leaves you a little lower. At three times, that «little lower» turns into a hole.
Repeat that back-and-forth twenty times and you'll see why this matters.
July 2026. Real prices, not an estimate:
| The chip index | −21,20 % |
| The 3X product | −56,99 % |
Let's compare with the figure being repeated. The index didn't fall 25 %, it fell 21.20 % — and that's debatable, because there are several ways to measure «the chip sector» and none gives the same number.
But the leveraged one didn't fall 45 %. It fell 56.99 %. That figure leaves no room for nuance: it's the closing price of a specific product. That's twelve points of loss the repeated figure leaves out.
And what matters isn't that a video gets it wrong. It's that someone hears «triples», sees «−45 %» and forms an idea of the risk that looks nothing like what can actually happen to them.
Here comes the part that usually goes untold, and which we think is the most useful part of the report.
When we measured how much it really multiplies, we got a spectacular number: in the months when the market fell, the product multiplied by five, not by three. A neat headline. We were about to write it.
It was false, and the mistake was ours. That «times five» came from dividing by tiny numbers.
If the market falls 0.4 % and the product falls 2 %, the division gives «times five». But nothing actually happened there: they're two minuscule moves. Put hundreds of months like that into the same average and the average goes haywire. The number was real; the conclusion, garbage.
Split it by the size of the move and the truth shows up, and it's more interesting than the headline we almost published:
| When the market… | Times measured | How much it really multiplies |
|---|---|---|
| falls more than 15 % | 74 | x2.69 |
| falls between 8 % and 15 % | 162 | x3.02 |
| falls between 3 % and 8 % | 354 | x3.43 |
| moves less than 3 % | 635 | this is where it breaks |
| rises between 3 % and 8 % | 612 | x2.72 |
| rises between 8 % and 15 % | 433 | x3.04 |
| rises more than 15 % | 171 | x3.27 |
In plain terms: when the market really moves —up or down—, the product does more or less what it promises. Around three times.
Where it hurts you is when the market goes nowhere. There it multiplies nothing: it bleeds you slowly while you watch the index and think nothing is happening.
That's the sentence we take away: the problem with these products isn't the crashes. It's the boring months.
What matters in this report isn't the product: it's how the name is set up. It's called «3X» and it multiplies by three each day's move, not the year's result. They're two different things, and the difference between them is kept by the product.
That isn't one particular firm's trick: it's how all products of this kind work, and it's written in their documentation. The problem is that the name is read in a second and hardly anyone opens the documentation.
What you take away is a cheap habit that saves a lot of money: before buying anything, find out what exactly it multiplies and how often. If the answer is «every day», you already know that a year out the result won't look like what the name promises. It applies to this and to any product with a number in front.
Good question, and it's worth answering before anyone asks it. Almost everything here was measured in a market that was rising.
Of the 2,462 sessions studied, 80.5 % were bull-market sessions —the index above its average of the last 200 days— and only 19.5 % bearish. Ten years in which chips were among the best things you could own.
And what changes depending on the market? Less than you might think:
| Situation | When the market rises sharply | When the market falls sharply |
|---|---|---|
| Bull market | x3.10 | x2.93 |
| Bear market | x3.03 | x2.98 |
In the big moves, the multiplier is more or less three in every situation. What changes radically isn't the multiplier: it's how much time you spend in the boring months. And bear markets have many more of them.
Here's the number that sums up the bear market, and this one doesn't depend on any odd division:
If someone had held this product only during the bear-market periods of these ten years, and been out the rest of the time, they would have lost 99.5 % of their money. The index, over those same stretches, lost 69.5 %.
And the other way round: whoever had been in only during the bull stretches would have multiplied their money absurdly. Neither is possible, because nobody knows which of the two they're in until afterwards. But it shows what everything depends on.
In a bull market. As of 4 August 2026, and despite the July scare, the index is still 33.9 % above its 200-session average. So far in 2026 there hasn't been a single bear-market day: zero out of 147.
Put another way: July was a correction within a rise, not a change of cycle. And in 2026 the product is up +196 % against the index's +73 % — again, around three times.
That last point describes what has happened, not a forecast of what will happen. If the market switches sides, we'll measure it again and publish it even if it makes us look bad.
| If you hold it | Times measured | The market did | Your 3X did | You lose in… |
|---|---|---|---|---|
| One month | 463 | +0,18 % | −2,17 % | 65 % |
| Three months | 233 | +0,21 % | −7,58 % | 94 % |
| One year | 74 | −0,23 % | −34,67 % | 100 % |
A whole year with the market stuck in the same place, and a third of the money has evaporated. In all 74 cases measured, without a single exception. You don't need to be wrong: being right and waiting is enough.
| From 2021 to 2025 | |
|---|---|
| The chip index | +152,6 % |
| The product that «triples» it | +46,8 % |
Five excellent years for the sector, and the leveraged product made less than a third of what simply buying the index made. It's not a rare case: it's what happened to anyone who got in in 2021 and held on.
Over the full ten years, with a huge and fairly sustained rise, the product made +8,817 % against the +5,796 % that «three times» would have been. It won, and by more than three times.
It's the same mechanism, working in your favour. When there's a trend and little back-and-forth, daily compounding pushes upward with the same force with which it sinks you in sideways months. The multiplier isn't three: it depends on the path the price takes to get there.
Correction — 18 August 2026
The first version of this report counted the first prediction in the table above as fulfilled —«in sharp falls it will multiply between x2.4 and x3.2»— with an «average x2.87, all within the range» and a hit tick. When we measured it again with the same data and the same cut-off (4 August 2026), neither the average nor the «all» holds up.
The average multiplier over the 74 months in which the index fell more than 15 % is x2.79, not x2.87. And nine of those seventy-four fall outside the stated range: six above x3.2 —up to x3.81— and three below x2.4, down to x2.24.
What does hold up, and is more useful than the original sentence: eight of those nine are the same episode, the crash of March and April 2020. The ninth (April 2025) falls outside by very little. In other words, the prediction worked in normal conditions and broke in the panic — and it broke on both sides: on the worst days the product multiplied by almost four, and on others in those same weeks it fell short of the forecast. When the market falls apart, the multiplier stops being reliable exactly when it's needed most.
The tick on that row goes from hit to partial. We corrected it because a prediction written before looking is only worth something if it's scored afterwards with the same yardstick it was written with.
We started this study with data ending on 30 December 2025 and without the real product —we used a basket of ten companies as a stand-in—. With that we wrote down four predictions. Then we got the data up to August 2026 and the real product. Result:
| What we said | What came out | |
|---|---|---|
| In sharp falls it will multiply between x2.4 and x3.2 | average x2.79. Within the range 65 of 74; the nine outside are the March-April 2020 crash, except one | ~ |
| It will return less than «three times» in ~7 out of 10 months | 65.1 % of 126 windows | ✓ |
| If the index fell ~18 %, the product fell between 44 % and 57 % | index −21.20 % → product −56.99 % | ✓ |
| In flat quarters it will lose money | in 2026 there was no flat quarter | — |
Our stand-in was terrible at calculating amounts. The simulated basket said that between 2021 and 2025 the product would have made +555 %.
It made +46.8 %. We were off by a factor of twelve.
The reason, now with the data in front of us: our basket gave the same weight to the companies that soared and to those that sank, and that looks nothing like how the real index is built.
The curious thing is what it got right anyway. The simulation said a year with a flat market would leave the product at −34.31 %. The real product gave −34.67 %. Thirty-six hundredths of difference.
A model can be bad at calculating amounts and good at understanding a mechanism. Confusing those two things is how many accounts get emptied.
Underneath all this there's nothing about chips or ETFs. There's a very human confusion: believing that what happens each day adds up by the end of the month. In the markets almost nothing adds up. It multiplies. And multiplying punishes back-and-forth and rewards trend.
That's why the same product can gain 8,817 % in ten years and, at the same time, lose to the plain market in four out of ten three-year periods. They're not two different products: it's the same one, on two different paths.
So when someone explains something to you with a verb —triples, protects, replicates, guarantees—, the question isn't whether it sounds good. It's: triples what, and how often?
We're people who measure. We got it wrong twice within this very report —with the stand-in that was off by a factor of twelve, and with the «multiplies by five» we nearly published— and both are told above, in full detail.
We'd like you to check these numbers. Seriously. The data are public closing prices and the calculations are written so anyone can repeat them. If you find a mistake, a period that doesn't fit or an interpretation we missed, write to us and we'll correct it right here, in public and with your name on it if you want it to appear.
We don't publish to be right. We publish to be right more often, and for that we need more eyes than ours. A method nobody can break isn't a method: it's a belief.
hola@sophronepsis.com — everything is read and answered.
Sophronepsis · Wait. Observe. Execute.
This report is about a product. But the decision that ruins the most people isn't which one you buy: it's how much.
The tool that calculates the how much. How much to risk on a trade isn't a hunch, it's a calculation. Free and without signing up.
Size the position without leaving the chart →
And the whole method, if you want to see it from the inside. Six days, one a day, with the numbers in front of you — including the ones that go wrong.
Start the walkthrough →
And if you don't want this to stay a snapshot. A report is a measurement from one day. Inside the platform the follow-up does not stop: measurements are redone when new data arrives, each strategy comes with its test and its published failures alongside, and the mentor answers whatever you ask about this at any hour. Subscribe and follow it from the inside: sophronepsis.com/mentoria.html
Measured on 5 August 2026 over 2,661 adjusted daily closes, from 4 January 2016 to 4 August 2026. The split by market type uses the 200-session average and covers 2,462 of those sessions. The predictions in section 6 were written with data ending on 30 December 2025 and checked afterwards. Where a figure couldn't be confirmed, we say so.
Sophronepsis — educational content, not financial advice. It is not a recommendation to buy or sell anything. Leverage multiplies the loss, not just the gain.