Trading systems research · Our own capital
Quantic Eagle builds trading systems: programs that turn market analysis into decisions about what to buy or sell and how much. We trade only with our own capital; we do not raise or manage other people’s funds. Here we share cases that help you understand and check those decisions.
Invented example · Not Quantic Eagle results
You are running a backtest: simulating a strategy’s buys and sells on past prices. You want to try a filter, an extra rule that skips some buy signals. Does it actually help you choose better trades?
Take every signal from the strategy, without the added filter. Use the same amount for each purchase.
The filter skips 12 trades. Expecting the remaining signals to be more reliable, you also choose to invest twice as much in each. Rerun the test over the same period.
These are separate tests; you are not changing positions that are already open. Both start with a simulated €10,000 account. In the second test you use more of that money per trade, without adding money to the account. At a price of €100, for example, you buy 20 shares instead of 10.
You made €154 more. Was that because of the filter, or because you doubled the trade size? The second test changed both which trades you took and how much you invested in each.
To measure the filter’s effect, rerun the second test with €1,000 purchases. The second button shows this comparison: identical prices and purchase amounts, with only the filter switched on or off.
Accumulated net profit · euros
Each point is a test: with or without the filter, buying €1,000 or €2,000 per trade.
Final net profit · €
A higher point means a larger final profit. A point further left means a smaller maximum drawdown: the account lost less money from a previous balance high. Here we measure the balance after each sale; temporary losses while a position is open are not visible.
Squares: no filter. Circles: filter applied. The number beside each point is the amount invested per trade in that test. The two filled symbols correspond to the comparison selected with the buttons.
You are comparing 48 trades at €1,000 without the filter against 36 at €2,000 with it. Profits are €294 and €448, but this comparison mixes two changes. Choose “Change only the filter” to see both versions buying €1,000 per trade.
Accumulated net profit · euros
Each point is a test: with or without the filter, buying €1,000 or €2,000 per trade.
Final net profit · €
A higher point means a larger final profit. A point further left means a smaller maximum drawdown: the account lost less money from a previous balance high. Here we measure the balance after each sale; temporary losses while a position is open are not visible.
Squares: no filter. Circles: filter applied. The number beside each point is the amount invested per trade in that test. The two filled symbols correspond to the comparison selected with the buttons.
Rerunning the filtered test with €1,000 purchases produces €188, versus €294 for the original strategy. The filter therefore cuts profit by €106 in this example. It also reduces maximum balance drawdown from €88 to €82: €6 less. You can now judge that trade-off; the €448 profit did not establish that the filter increased profit.
This can happen when you try a new strategy version: you add a rule and adjust position sizes too. To find out which change helps, first compare versions with one thing changed at a time.
| Strategy | Amount per trade | Net profit | Maximum balance drawdown |
|---|---|---|---|
| Without filter | €1,000 | +€294 | €88 |
| With filter | €1,000 | +€188 | €82 |
| Without filter | €2,000 | +€684 | €168 |
| With filter | €2,000 | +€448 | €142 |
At €2,000 per trade, you can also compare both strategies: profit is €684 without the filter and €448 with it. The filter therefore reduces profit by €236. Why is that not twice the €106 difference in the €1,000 tests? Doubling the amount doubles gross gains and losses, while the cost stays at €2 per trade. Skipping 12 trades always saves €24 in costs. These are this example’s assumptions, not a general law about filters or AI models.
Maximum balance drawdown is the largest fall from a balance high to a later low. For example, if the balance reaches €10,100 and then falls to €10,000, the drawdown is €100. Here we observe only the starting balance and the balance after each sale; the account could have lost more while shares were still held.
Each test starts with €10,000. Trades are sequential: sell the entire position before opening the next one. Within each test, invest the same amount per trade, without leverage or increasing it as profits accumulate.
Every purchase is made at €100 per share. Sale prices and the 12 skipped trades are already fixed in the invented data. We are not searching for a profitable filter; these choices illustrate how to compare two versions. A completed trade costs €2 in total for buying and selling; a skipped trade has no cost or profit.
Equal amounts per trade do not guarantee equal risk: the filter changes which trades are taken. This example does not evaluate a real AI model or establish future profits. Other costs, differences between expected and actual execution prices, and order execution constraints are excluded.
node QE-COMPARE-02.js
Why compare every combination? NIST/SEMATECH explains experimental design and interactions between changes. It is a methodological reference, not a validation of this example.
QE-COMPARE-02 / 1.0 · 20.09.2026 · Quantic Eagle LTD
01 / Risk along the way
Two paths can finish at the same level after exposing capital to very different losses.
02 / Testing the assumptions
A promising result may depend on low costs or execution that is hard to achieve. Those assumptions need testing too.
Brand film · 49 seconds
Your program buys after a price fall. You want to add a rule that skips some of those purchases. Run two simulations over the same period, one with the rule and one without it. Keep purchase amounts and cost assumptions the same, then compare the trades, gains and losses. If you change the amounts too, the comparison no longer isolates the rule’s effect.
In an invented example, the program buys a stock if the company announces earnings above expectations. The announcement comes out at noon on Monday, but the simulation records the purchase at 10 am that day. The program used the news two hours before it was available. Compare the announcement time with the purchase time to find the error: a date marked “Monday” hides the difference.
In an invented example, buying and selling earns €5 before costs. Fees and differences between the assumed prices and those actually obtained cost €7 in total, leaving a €2 loss. In your own test, account for these costs on every trade before judging the final result.
In this invented example, you are comparing two versions of a program that simulates stock purchases and sales over the same month.
scores.csv. Each row contains a date, stock and score, including stocks you did not buy. With the same model and prices, you can use those numbers to test the new threshold. You must still simulate the orders again: A was previously excluded; the program can now buy shares in A and open a position, meaning it records those shares among the account’s holdings.Suppose the simulated account has €1,000 in cash. You buy 8 shares at €100 and pay a €1 fee. That leaves €199 in cash and a position of 8 shares. If the price stays at €100, the account is worth €999: €199 + 8 × €100. The €800 spent has become shares; the €1 fee is the cost. On selling, the program adds the proceeds to the cash available and subtracts the specified fee. Repeating these steps reconstructs the account under the new rule.
The model described above receives only prices, so the cash remaining does not affect its scores. A different model might be designed to receive available cash as well. The simulation program would then supply the number recorded in the account: €199 after this purchase, compared with €1,000 in the run without it. The model runs its calculation with the new inputs; you cannot automatically copy the score produced with €1,000. Calculating a prediction with different inputs does not mean retraining the model.
On day 15 you add a rule and run the simulation again from the start of the month. The new chart can help you understand the change. It does not show what the program actually decided during the first fourteen days, when it was still using the old rule.
The prices and predictions are made up for this exercise. They are not Quantic Eagle trades or results.
Clear boundaries
Quantic Eagle develops and uses its own programs to decide how to invest the company’s money. We carry out research, set buying and selling rules, and manage risk controls internally. We do not manage money entrusted to us by clients.
Is
Is not
Looking into joining an existing company or working with its founders? We’ve brought those articles together in a separate reading path.
Build, buy or back? A guide to what ownership means, what more capital changes and what risks remain.
Equity, capital, responsibilities and the questions to resolve before joining.
How to assess what happened and choose the next stage of an entrepreneurial career.
Separate the title from ownership, work, capital and authority.
Proposals for the company
We may consider proposals concerning the company’s work, such as technical collaboration, software, data or a role carrying responsibility within the business. Explain what you propose, what you can contribute and why you think it would help Quantic Eagle. Any question of share ownership requires a separate private assessment; this website does not offer shares to the public. We do not accept money to invest on other people’s behalf.
Company & strategic enquiriesCompany & strategic enquiries
Describe the project or reason for your enquiry and explain what you would like to discuss with the company. For personal-data requests or security reports, select the corresponding form option.