Somewhere between a large strategy library and a single, deployed, monitored portfolio, a lot of decisions have to be made — most of them not obvious from the outside. This article steps back from any single topic covered elsewhere in this series and lays out the conceptual journey as a whole: the stages a structured process moves through, and why each one exists.
1. Strategy library
Everything starts with a pool of candidate strategies — generated, researched, or imported from a tool such as StrategyQuant X. At this stage, the raw material is quantity and variety: more credible candidates give later stages more to work with. See From StrategyQuant X to a Portfolio for how this stage typically feeds into what follows.
2. Individual quality
Before any strategy is considered for a portfolio, it needs to hold up reasonably on its own — a credible sample size, sensible risk-adjusted metrics, and no obvious signs of curve fitting. This stage answers “is this strategy good?” on its own terms; see Profit Factor, Expectancy, Stability and Trade Count and Why a Great Backtest Does Not Automatically Mean a Great Trading Strategy.
3. Diversity
The surviving library is then examined for genuine behavioural variety — market, asset class, direction, timeframe, entry and exit style — not just symbol count. This is where the difference between “looks diversified” and “is diversified” starts to become visible; see Diversification in Algorithmic Trading Is More Than Different Markets.
4. Correlation
Where diversity looks promising on paper, checking historical correlation between serious candidates confirms whether the intended variety actually showed up in how the strategies performed — with the caveat that correlation is historical evidence, not a guarantee; see Correlation in Algorithmic Portfolios.
5. Portfolio construction
This is where individual strategies become a combination — and where the question changes from “is this strategy good?” to “does this strategy improve the portfolio it would join?” The best individually-ranked strategies don’t automatically produce the best combination; see Why the Best Individual Strategies Don’t Always Build the Best Portfolio.
6. Portfolio comparison
Rarely is there one obviously correct portfolio. Comparing several credible candidate combinations, side by side, on genuine combined-portfolio metrics — not averaged member statistics — surfaces trade-offs a single candidate viewed in isolation would hide.
7. Robustness
A single historical equity curve is one realised path through a wider space of things that could plausibly have happened. Stress-testing a candidate portfolio’s sensitivity to trade sequencing gives a more honest sense of the drawdown and return uncertainty involved than trusting the one historical path alone; see Monte Carlo Testing Explained.
8. Final analysis
Before deployment, a full drill-down into the chosen portfolio’s combined metrics, equity curve, membership, and diversification profile gives a last, detailed check — confirming the combination behaves the way the earlier stages intended, not just assuming it.
9. Deployment
Moving a chosen portfolio from analysis to a real, connected account. This is a genuine transition, not a formality — a backtest and a live account are related but distinct experiments; see Backtest vs Live Trading.
10. Live monitoring
The process doesn’t end at deployment. Comparing live behaviour against historical expectation — patiently, with enough evidence, across return, drawdown and trade frequency separately — is how a trader knows whether the portfolio is still doing what it was built to do; see How to Tell When a Live EA Is Behaving Differently From Its History.
Why laying it out this way helps
Seen as a whole, the point of this sequence isn’t that any one stage is more important than the others — it’s that skipping a stage, or collapsing several into one shortcut (picking “the best strategies,” say, and calling that a portfolio), tends to reintroduce exactly the risks the other stages exist to catch. A structured process is really just a way of asking each necessary question at the point where it can still be answered honestly, rather than discovering the answer live, with real capital.
Where this fits in the SIPS workflow
This sequence closely mirrors the real, implemented SIPS workflow: Understanding the SIPS Workflow covers how the stages connect end to end, Portfolio Builder and Portfolio Comparison cover construction and comparison, and Monte Carlo & Robustness covers the current state of the robustness stage. None of this discloses exactly how any specific search or ranking is performed internally — that detail is intentionally out of scope here, and elsewhere in SIPS’s public documentation.
The practical takeaway
Going from a hundred strategies to one trusted portfolio is a sequence of distinct, honest questions, not one big decision made all at once. Each stage in this article exists because skipping it tends to hide a specific kind of risk — concentration, overlap, curve fitting, sequencing luck — until it’s much more expensive to discover.