How to Backtest a Strategy on Seentio

Backtesting answers one question: "If I'd been running this strategy for the last few years, how would it have done?" The answer is never perfect — past performance never guarantees future results, and even our walk-forward methodology has limitations we'll be honest about. But it's the best directional signal available before you commit real capital.

This guide walks you through running a backtest on any strategy you've saved on Seentio.

What you'll learn

How to run a 1, 3, or 5-year walk-forward backtest on any saved strategy via Claude Code, what the output means, and what backtests don't tell you.

Prerequisites

• A saved strategy (see How to create a strategy)
• Either Claude Code with MCP installed, or access to the Strategies page on the web (both surfaces hit the same backend)
• 5 AI credits per backtest (Basic tier has 200/month free, plenty for a few dozen runs)

Step-by-step walkthrough

1. Identify the strategy you want to test

Ask Claude:

What strategies do I have?

Claude calls list_strategies and shows you a table with name, status (draft/active/paused), and creation date. Pick the one you want to backtest. If you've just created a Magic Formula, that's the one.

2. Run the backtest

Backtest 'Greenblatt Magic Formula' for 3 years against SPY.

Claude calls backtest_strategy with your strategy's full definition, period_years=3, and benchmark="SPY". This call takes 3-8 seconds typically — we're fetching daily price history for up to 30 stocks plus the benchmark, then walking forward through ~750 trading days.

3. Read the output

You'll get back something like this:

Plus a rebalance log showing the most recent 3 rebalance events: which tickers were added, which dropped, the order quantities. Useful for sanity-checking that the strategy actually did what you expected.

4. Compare against the benchmark

The benchmark is your "do nothing different" baseline. SPY (S&P 500), QQQ (Nasdaq 100), IWM (Russell 2000), and DIA (Dow 30) are the four supported choices.

Some questions to ask:

• Did your strategy beat the benchmark? Annualized return is the headline number. A 5%+ outperformance over 3 years is meaningful; less than 2% is noise.
• Did it do so with similar risk? Look at Sharpe ratio (higher = better risk-adjusted return) and max drawdown (less-negative = lower worst-case loss). A strategy that beats SPY by 5% but has a 40% drawdown is not necessarily a good trade.
• Is the win consistent? A strategy that's +50% one year and -10% the next has the same average as one that's +20% per year, but the second is much easier to live with. The volatility number captures this.

5. Decide what to do next

Three honest options:

• Activate if the backtest looks reasonable and you understand the caveats below: "Activate it on my ETrade, email alerts only."*
• Iterate if it doesn't: ask Claude to tweak the spec (different top_n, different exclusions, different cadence) and re-backtest. Each iteration costs 5 credits.
• Walk away if no version looks compelling. Save the strategy as a draft for later or delete it.

What backtests DON'T show you

Be honest with yourself about the limitations:

Survivorship bias

Our backtest uses today's universe snapshot for the entire historical period. A company that delisted 3 years ago is invisible to the backtest. If your Magic Formula would have picked a stock that subsequently went to zero, you don't see that loss — because the stock isn't in today's universe.

This biases historical returns upward. The bias is bigger for longer lookback periods and for strategies that pick from broader universes. Eliminating it requires loading the universe-as-it-was on each historical rebalance date, which is a feature on our roadmap.

Trading costs

Backtests assume you got fills at the official close price with zero commission and zero slippage. Real-world trading has spreads, market impact (especially for smaller stocks), and commission (though most retail brokers are zero-commission for stocks now). For monthly-cadence strategies on liquid large-caps, this is small. For weekly strategies on small-caps, it's not.

Behavioral risk

The backtest shows what would have happened if you'd held the strategy through every drawdown without flinching. Real investors flinch. A backtest with -18% max drawdown means at some point you'd have been down 18% from the peak — would you really have stayed the course?

The future is not the past

The most important caveat. Strategies that worked historically can stop working. Factor strategies (value, momentum, quality) all go through extended periods of underperformance. The Magic Formula spent 2018-2020 underperforming SPY before recovering. Backtests show what the past looked like — they cannot predict what the next 5 years will look like.

Tips & common questions

• Run multiple periods, not just one. A 1-year backtest can be dominated by recent narrative-driven moves; a 5-year backtest tells you whether the strategy works across regimes (rising rates vs falling rates, growth vs value rotations). Compare 1y, 3y, and 5y outputs side by side.

• Read the rebalance log. It tells you which tickers actually drove the returns. If 80% of the gain came from one or two names, the strategy might just be a momentum bet on those names rather than a real factor strategy.

• Don't over-optimize. It's tempting to tweak parameters until the backtest looks perfect. That's curve-fitting, and it doesn't generalize. If a strategy needs 7 specific parameters to look good, it's probably not a real strategy.

• Compare to a simpler alternative. If your fancy multi-factor strategy barely beats top_n_large_cap (the boring buy-the-biggest-stocks template), the complexity isn't earning its keep.

• The rebalance count matters. A monthly-rebalance strategy over 3 years should have ~36 rebalances. If yours has 5, something's wrong with the cadence config — ask Claude to check.

What's next

• Create your first strategy using Claude Code if you haven't already.
• Once you activate, learn how the alert system fires when monthly rebalances are due.
• Understand what "live strategy" means before connecting your brokerage.

Need more help? Visit our Chat page to ask Seentio's AI assistant, or reach out at contact@app-seentio.com.