Algorithmic trading tools

1. A platform to execute on

MetaTrader 4 and 5 for Expert Advisors, cTrader for C# strategies, or a broker's API for a bot that runs on a server. All free to use — the cost is in the trading, not the platform.

Choose on whether the bot must run unattended. MetaTrader needs its terminal open, so continuous operation means a VPS; an API bot runs on a server by design.

2. Market data

Your broker supplies live prices and historical candles, usually free through the platform or API. That is sufficient for most price-based strategies.

Paid data services matter when you need something the broker does not provide — tick-level history, a longer archive than the platform keeps, or data from markets you do not trade. For a bot trading one broker's instruments, the free feed is normally the right feed, since it is also the one you will execute against.

3. A backtesting engine

MetaTrader has one built in and it is good, with multi-threaded optimisation. Outside MetaTrader, backtesting generally means writing it or using a Python framework.

The requirement that matters more than the tool: it must charge spread and commission, and it must not let the strategy see data from bars that have not closed. An engine that gets either wrong produces confident, useless numbers.

4. An optimiser, used carefully

Parameter search is the most dangerous tool in the set, because it will always find something. Keep grids coarse — fine grids find sharper peaks and sharper peaks are more likely to be artefacts — and judge candidates only on data they were not fitted to.

Then check the neighbourhood of any winner. If nearby parameter values also work, you have a plateau; if the winner is alone, you have a coincidence that will not repeat.

5. Risk controls

Not an optional extra, and not something to bolt on afterwards. A stop loss on every position placed at the broker, a daily loss cap, a limit on trades per day, a cap on open positions, and an automatic pause after a bad run.

The design point worth getting right is separating an automatic pause from a kill switch that waits for a human, and putting the pause threshold below the kill switch. Get that the wrong way round and the system needs attention several times a day while claiming to be automated.

6. Monitoring

A dashboard showing profit factor with its trade count, expectancy, drawdown and the current losing streak — plus a live feed of what the bot is doing and, when it declines a signal, which limit refused it.

That last item is the one most systems lack and the one that matters most day to day. A bot generating signals and taking no trades is either correctly protected or silently broken, and only the refusal reason tells you which.

What you do not need

• Paid signal services. If the signal were reliable it would not be for sale at that price.
• Machine learning, until a simple rule-based strategy has been made to work. A complex model on a market with no edge produces a more elaborate way of losing.
• Expensive data, for a bot trading one broker's instruments on price alone.
• A faster VPS, unless you have measured latency and found it to be the problem. It usually is not.

Common questions