How to choose automated trading software

How to choose automated trading software: eliminate before comparing

Most buyers start by comparing features across a shortlist, which is the expensive way round. Start instead by eliminating on architecture. Four questions remove most of the market in an afternoon, and what survives is worth the detailed comparison that follows.

This page is the checklist we would use. It is deliberately ordered, because the early questions save you the trouble of asking the later ones.

The four elimination questions

Where does the decision happen

On your machine with your keys, or on a vendor's servers. If it is theirs, then their outage, their breach and their business continuity become risks to your account, and you have no way to inspect either. This is the fastest way to shorten a list.

Where does protection rest

A protective order held at the brokerage survives the software crashing, the machine rebooting and the vendor disappearing. One held inside the software does not. Ask the question plainly and be wary of an answer about how stable the software is.

What record does it keep

You want a record of decisions, including refusals, that you can read afterwards. Without it you cannot audit, you cannot improve, and you cannot explain to anyone else what happened and why.

Can you reproduce the research

Ask whether you can rerun the backtest yourself, on your machine, from one command, with a real cost model in it. A vendor who cannot offer that is asking you to take the numbers on faith.

What to examine on the survivors

The evaluation terms

A real evaluation runs on a paper account for long enough to see more than one kind of market. Two weeks shows you nothing. Ask how long, on what account, and whether the evaluation is the full product or a reduced version.

The license scope

Per operator, per machine, per account and per entity price very differently. Establish the unit before the number, and get it in writing, because this is where disputes start.

The support path

Establish who you reach, how, and how quickly, when something misbehaves outside business hours. Trading software fails at inconvenient times by definition.

What updates mean

Whether updates are included, whether they can change behaviour without your approval, and what happens if you decline one. All defensible answers, but you should know which you are getting.

Deciding, and what to check afterwards

Run it in paper past the point of boredom

The temptation is to go live as soon as it looks good. Wait until you have seen it refuse trades, handle a bad data day and survive a reboot. That takes months, not weeks.

Reconcile against the broker before you trust the screen

The software's own view of your positions and profit should be rebuilt from the broker's record, not from what it intended to do. Check that this reconciliation exists and that it runs before new orders.

Size the first live deployment to be survivable

Whatever you conclude from paper, the first live capital should be an amount whose total loss would be annoying rather than serious. Scale on evidence from your own account, not on the vendor's.

Keep the record and read it

The point of a decision record is that someone reads it. Set a cadence, look at the refusals as well as the trades, and be willing to stop.

Answers that should end the conversation

We cannot show you the decision record

If a system cannot produce a readable account of what it decided and why, nobody can audit it, including the vendor. You are being asked to trust an outcome with no way to examine the process that produced it.

The stop is handled in software

This means an unprotected position every time the software is not running, which includes crashes, updates, reboots and power cuts. It is the single most consequential shortcut a vendor can take and it is often not volunteered.

Our returns speak for themselves

Returns are the easiest number to select favourably. A vendor who leads with them and cannot produce a reproducible backtest with a real cost model is selling a curve, not a system.

You do not need to see the charter

Whatever a vendor calls the document that says what the system will never do, you should be able to read it before buying, not after. A refusal here tells you the constraints are informal, which means they can change.

Common questions

What should I look for in automated trading software?
Start by eliminating on four things: whether decisions are made on your machine or a vendor's servers, whether protective orders rest at the broker or only in software, whether it keeps an auditable record including refusals, and whether you can reproduce its backtests yourself. These remove most of the market before you compare features.
How long should I test automated trading software before going live?
Long enough to see more than one kind of market, which usually means months rather than weeks. You are waiting to watch it refuse trades, handle a day of bad data and survive a reboot cleanly. Going live after a good fortnight tells you almost nothing about how it behaves when conditions change.
What is the most common mistake when buying trading software?
Comparing features across a shortlist before eliminating on architecture. Buyers spend weeks weighing indicator counts and interface quality between products that should already have been ruled out because the vendor makes the decisions on their own servers or keeps protection only in software.
What are the warning signs when buying automated trading software?
Four answers should end a conversation: that no decision record is available for you to read, that protective stops are handled inside the software rather than at the broker, that published returns are offered in place of a reproducible backtest, and that the document setting out what the system will never do is not available before purchase.

TradeAgentic is an autonomous trading desk for macOS and Windows, licensed to operators and firms who intend to run it themselves.

Request licensing terms