Next steps

You now have a complete, small, honest pipeline: data in, validated; indicators you wrote and tested; a vectorized backtester for search and an event-driven one for realism; costs, lag and ATR sizing; metrics and bootstraps; walk-forward and purged cross-validation; and a live path with a paper broker, logging, alerts, idempotent orders, state and kill switches. This final lesson is about what to do with it, what to study next, and what not to do.

What you have and what you do not

You have the machinery to test a trading idea honestly and to run it carefully. You do not have a trading idea that works, and nothing in this course claimed to give you one. The moving-average crossover and the z-score reversion were teaching vehicles, chosen because they are simple enough to verify by hand, not because they are edges. On most liquid instruments over most periods, after costs, simple rules of this kind produce results indistinguishable from noise in a walk-forward, and the pipeline you built will tell you so.

That is the right outcome. The alternative, a pipeline that flatters ideas, is the thing that loses money.

Where edges tend to come from

Not from indicators. The people who make systematic trading work over years usually get their edge from one of:

• Structural effects: rebalancing flows, index inclusion, options expiry, futures roll, funding rates, earnings drift. Things that happen because of how markets are organised, which are documented, and which persist because the participants causing them are not trying to avoid them.
• Breadth: a small effect applied across many uncorrelated instruments, which turns a t-stat of 1 on one symbol into a t-stat of 3 on nine.
• Execution quality: earning the spread rather than paying it, which requires intraday data and infrastructure beyond this course.
• Patience: slow strategies that trade a few times a year, survive costs, and are unpleasant to hold, which is why they persist.

Whatever you test, the pipeline stays the same. Only the signal function changes.

The two courses to take next

Risk Management is the companion to this one, and if you have not taken it, take it now. Everything in Module 5's sizing and Module 9's kill switches assumes you have a written risk plan with a per-trade risk, a position cap, a daily loss limit, and a drawdown at which you stop. The plan is what the code enforces; without it, the code is enforcing numbers you made up in a config file at midnight.

Backtesting and Strategy Validation goes deeper into Module 8: deflated Sharpe ratios, the full combinatorial purged cross-validation, multiple-testing corrections, regime analysis, and how to build a research log that keeps you honest across months of experiments. This course gave you the intuition; that one gives you the statistics.

Tools to graduate to

Nothing in this course used a backtesting library, deliberately, so that every number came from code you could read. Once the concepts are solid, the libraries save time:

• vectorbt for fast vectorized research across large parameter spaces.
• backtrader or Nautilus Trader for event-driven backtesting with many order types and multiple instruments.
• QuantStats for tear sheets with the metrics from Module 6 and many more.
• Polars as a faster alternative to pandas when minute data gets large.

Use them the way you would use a calculator after learning arithmetic: to go faster, not to skip understanding. When a library's number disagrees with yours, that is a learning opportunity, and it will happen.

Reading

• Robert Carver, Systematic Trading: how a former institutional systematic trader thinks about rules, sizing and combining systems. The best single book for this course's audience.
• Marcos López de Prado, Advances in Financial Machine Learning: the source of purged cross-validation, deflated Sharpe, and a bracing view of how most backtests are wrong. Dense; read chapters 7, 11 and 12 first.
• Ernest Chan, Quantitative Trading: an older, practical introduction with the same emphasis on costs and realism.
• The documentation of your broker's API, read in full, once. It is the least exciting and most valuable reading on this list.

What not to do

• Do not go live because a backtest looked good. Go live because a walk-forward looked acceptable, the bootstrap drawdown fits your plan, a month of paper trading matched the backtest, and the review checklist has no failures.
• Do not size up because the first month worked. The first month is one draw from a distribution you sketched in Module 6.
• Do not add parameters to fix a bad backtest. Each one is a degree of freedom for noise.
• Do not run anything with real money on a laptop that sleeps.
• Do not skip the daily "ok" alert. Silence must mean failure.

Key idea: The pipeline is finished; the research is not. Keep the machinery fixed and honest, vary only the idea, log every trial, and let the walk-forward say no as often as it needs to.

A last exercise

Pick one structural effect from the list above that applies to a market you already trade by hand. Write the rule as a sentence. Turn it into a signal function with the interface from Module 4. Run the full project script on it with real data at base and double costs. Produce the report card and the eight-item report. Then post it, including the failures, in the community. The traders who improve fastest are the ones whose "no edge" reports are as public as their wins.

Try it: Set a recurring reminder for one month from today. When it fires, open the trials log and count. If the number of trials is large and the number of ideas that survived walk-forward is zero, you have been doing it correctly. If the number of trials is small and something survived, look for the leak before you look for a broker.

Recap

• You have the machinery to test and run ideas honestly; the ideas themselves are the ongoing work.
• Edges tend to come from structure, breadth, execution and patience, not from indicator settings.
• Take the Risk Management course now and the Backtesting course next.
• Graduate to libraries for speed once you can reproduce their numbers by hand.
• The pipeline stays fixed; only the signal changes; every trial is logged; walk-forward gets the final word.