Guide & Methodology
This is the canonical operating manual for Alphrex: workflow, metrics, methodology, data sources, and trust limits in one reference surface.
Start here
What are you trying to do?
Pick a path — or scroll down for the full reference.
Your starting point. Browse your watchlist, see what's moving today, and pick one ticker or theme to investigate.
Find one ticker or theme worth testing — don't act on anything here.
Use the Market Snapshot to identify which assets are moving and why.
When you've picked a ticker, move to Strategies.
Pick the rule you want to test. Choose a built-in strategy and click Backtest — or use Strategy Studio to write your own idea.
Start with a Beginner strategy for your first run.
Read the signal logic description before running — understand what you're testing.
The idea needs to be specific enough to express as a rule.
Run the strategy against historical data. Set realistic costs, use a 20% OOS split, and read OOS Sharpe first — not the headline chart.
Use at least 5 years of data and set Commission to 5 bps, Slippage to 3 bps.
Set OOS Split to 20% — this is the only honest performance test.
If OOS Sharpe is negative, the strategy failed — don't promote it.
A good backtest only proves the strategy didn't fail in the past. Paper tracking lets you watch it forward in real time — no real money.
Only promote after IS Sharpe ≥ 1.0 and OOS Sharpe ≥ 0.5.
Start with a long-only sleeve — short exposure should be tested first.
Wait 60 live days before drawing any conclusions.
Markets change and so do strategies. The signal monitor shows live signals daily — use it to decide when to keep watching, or when to stop.
Filter to LONG or SHORT with min strength ≥ 0.5 before acting.
Always backtest a signal on the specific ticker before treating it as actionable.
Retire a strategy when live Sharpe drops below 0.3 for 30+ consecutive days.
Common questions
How to: Run your first backtest
Backtesting applies a strategy's entry/exit rules to historical price data to show how it would have performed. This is your starting point before any other feature.
1
Pick a strategy
Go to /strategies and choose one at Beginner difficulty — try Momentum or RSI Mean Reversion. Click Backtest on its card to pre-fill the form.
2
Set ticker and date range
Enter SPY for your first run. Set the date range to at least 5 years — shorter ranges have too few independent market regimes to say anything meaningful.
3
Set realistic costs
Commission: 5 bps. Slippage: 3 bps. Never run at 0 bps — a strategy that only works at zero cost has no real edge and will fail live.
4
Set OOS Split to 20%
This reserves the last 20% of your date range as an untouched holdout. Enable it every time — it is the only honest performance test.
5
Read OOS Sharpe first
Ignore the headline chart. OOS Sharpe >= 0.5 is credible; OOS Sharpe < 0 means the strategy failed the honest test. IS Sharpe will always look better — that is expected.
Demo — Step 1
OOS Sharpe >= 0.5 = credible edge
How to: Set up a paper sleeve
Paper trading creates a forward record from today — no real capital, just live signals. Only promote a strategy here after it has passed a backtest with realistic costs.
1
Verify your backtest first
Check: IS Sharpe >= 1.0, OOS Sharpe >= 0.5, IS/OOS gap <= 0.5, and realistic costs (>= 5 bps commission). If OOS Sharpe is negative, stop — the strategy is not ready.
2
Go to /paper and click Add Strategy
Pick the same strategy and ticker you backtested. The system freezes today as the live start — any earlier history only warms up the signal.
3
Choose sleeve type
Long-only: follows LONG signals only. Suitable for most equity strategies. Long/short: also goes short — only use if your backtest explicitly modelled short positions and they contributed to Sharpe.
4
Wait 60 live days
Scoreboards require a minimum of live days before ranking. Treat anything under 30 days as noise. A clean signal needs ~60 days of forward data before you can draw conclusions.
Demo — Step 1
Track live signal quality from today's date
How to: Compare two strategies
The compare page runs multiple strategies on the same ticker and overlays equity curves. Use it to check if your best backtest is genuinely different — or just the lucky pick from a large set.
1
Go to /compare
Navigate to the Compare page from the sidebar.
2
Select 2-6 strategies from different categories
Pick at least one momentum, one mean reversion, and one volatility strategy. Testing only one category cannot distinguish genuine edge from regime luck.
3
Use the same settings for all
Apply identical ticker, date range, and costs (5 bps commission, 3 bps slippage, 20% OOS split) to all strategies — you are comparing on equal footing.
4
Check curve divergence
Curves that converge tightly are trading the same underlying bet — no diversification benefit. Genuine alternatives diverge, especially during regime transitions.
5
Sort by OOS Sharpe, not CAGR
The metrics table shows OOS Sharpe for each strategy. High CAGR often just means more risk. OOS Sharpe is the ranking that matters.
Demo — Step 1
Sort by OOS Sharpe - not total return
How to: Read Sharpe and OOS
IS Sharpe and OOS Sharpe tell you most of what you need to know about a backtest. Here is how to read them without fooling yourself.
1
IS Sharpe — upper bound only
The headline Sharpe over the full date range. > 1.0 is acceptable, > 1.5 is strong. But IS Sharpe is always inflated — treat it as a ceiling, not a target.
2
OOS Sharpe — the honest test
Sharpe computed only on the holdout period (last 20%). This is the number that matters. >= 0.5 is credible; < 0 means the strategy failed.
3
IS/OOS gap
Subtract: IS Sharpe minus OOS Sharpe. < 0.3 = clean (low overfitting). 0.3–0.5 = acceptable. > 0.5 = red flag. > 1.0 = likely curve-fitted, do not trade.
4
Walk-forward Avg Sharpe
Average Sharpe across walk-forward test windows. If much lower than IS Sharpe, the strategy learned a specific regime — not a general pattern.
Demo — Step 1
Read IS, OOS, and walk-forward Sharpe together
How to: Read today's signals
The signal monitor shows the current LONG/SHORT/FLAT position for every strategy-ticker pair. Signals update once per day at 18:00 UTC from daily closing prices.
1
Go to /signals
Navigate to the Signal Monitor from the sidebar.
2
Filter by direction
Use the Direction filter to show only LONG or SHORT. FLAT is the default state — most pairs will be FLAT most of the time.
3
Set min strength >= 0.5
Strength is a normalised [0,1] conviction score. Below 0.5 the signal is weak and likely to flip. Above 0.7 is high-conviction.
4
Backtest before acting
Click Backtest on any signal row to verify the strategy's history on that specific ticker. A LONG signal is a hypothesis to test, not a buy recommendation.
5
Check the last updated time
If signals have not refreshed since your target market close, they may be stale. Trigger a manual refresh if needed — auto-refresh runs at 18:00 UTC.
Demo — Step 1
A signal is a hypothesis - always backtest before acting
How to: Build a paper portfolio
Portfolios aggregate multiple paper sleeves with custom weights, letting you track a blended allocation rather than individual strategies.
1
Create at least two paper strategies first
Go to /paper and click Add Strategy to set up two or more paper sleeves. You need at least two to form a portfolio.
2
Go to the Portfolios tab
On the /paper page, click the Portfolios tab at the top.
3
Click New Portfolio
Give it a name. You will then add component strategies and assign weights.
4
Add components and set weights
Add each strategy and assign weights. Weights must sum to 100%. Start with equal weights (e.g. 50%/50% for two strategies).
5
Adjust after 60+ live days
Portfolio NAV is the weighted sum of component returns. Equal weighting is the least-biased starting point — only adjust when you have 60+ live days of data per component.
Demo — Step 1
Build allocation weights only from sleeves you already track
What tickers can I use?
The Free plan has a reliable core universe of liquid US tickers. Start there for the cleanest first backtests.
1
US ETFs always work on Free
SPY, QQQ, IWM, GLD, TLT, EEM, XLF, XLE, VNQ, DIA, HYG, LQD — these are the safest tickers for first-time backtests.
2
Large-cap US stocks generally work
AAPL, MSFT, GOOGL, AMZN, NVDA, META, TSLA, JPM, BRK-B, JNJ, V, WMT — widely available with clean daily data.
3
Avoid non-US suffixes on Free
Tickers with country suffixes (.L, .DE, .T) require Pro. On Free, stay inside the core US universe.
4
If a backtest fails with a data error
Try SPY first to confirm the strategy itself works. If SPY runs fine, the original ticker is outside the reliable contract. Swap for a US equivalent.
5
See section 04A for the full list
Scroll to the Free-Plan Core Tickers section below for the complete validated list, organised by asset class.
Demo — Step 1
Start with the validated Free-plan core universe
How to: Avoid overfitting
Overfitting means the strategy learned the noise in the training data, not a real pattern. It is the main reason backtests that look great fail in live markets.
1
Always set OOS Split to 20%
If a strategy cannot survive 20% of its own data being held back, it has no business being paper-traded. Never skip the holdout.
2
Check the IS/OOS Sharpe gap
IS minus OOS > 0.5 is a red flag. OOS Sharpe < 0 means the strategy outright failed. Do not rationalise these numbers — they are telling you something real.
3
Run walk-forward validation
Set Validation Folds to Sequential WF, 5 folds. Consistent Sharpe across folds is a stronger signal than high average Sharpe with extreme variance.
4
Compare against multiple strategies
Use /compare with 3-4 strategies from different categories. If yours is the only one with good OOS Sharpe, it may be the lucky draw from a large search.
5
Use realistic costs throughout
Overfitting is most obvious at 0 bps but invisible at 5 bps. If OOS Sharpe drops to negative when you add 5 bps commission, the apparent edge was entirely in transaction timing.
Demo — Step 1
A wide IS/OOS gap is evidence against the edge
How to: Write a custom strategy
Strategy Studio lets you define your own signal logic in plain terms. Use it when the built-in library doesn't capture your idea — no code required.
1
Go to /strategies and click Strategy Studio
Strategy Studio is available from the Strategies page. It lets you describe signal logic in structured plain English rather than raw code.
2
Describe your entry condition
Be specific: name the indicator, the threshold, and the direction. Example: Buy when the 14-day RSI crosses below 30 and price is above its 50-day moving average.
3
Review the generated parameters
Strategy Studio translates your description into tunable parameters. Check that the parameters match what you intended before running.
4
Click Backtest to test it immediately
The strategy is pre-filled in the backtest form. Set realistic costs (5 bps commission, 3 bps slippage) and a 20% OOS split before running.
5
Save to My Strategies if credible
If OOS Sharpe >= 0.5 and the IS/OOS gap is <= 0.5, save the strategy. Saved strategies appear in the library and can be used in Compare and Paper Trading.
Demo — Step 1
Choose a template, inspect parameters, then run the draft
How to: Diagnose a bad backtest
A poor backtest result usually has a specific cause. Here is how to identify which problem you are looking at and what to do about it.
1
OOS Sharpe is negative
The strategy failed the only honest test. Do not try to tune your way out — reduce the number of parameters, increase the OOS split to 30%, or switch to a simpler strategy.
2
IS Sharpe looks great but OOS is near zero
The IS/OOS gap is too large — the strategy learned noise, not signal. Run walk-forward validation to confirm. A consistent gap across folds means systematic overfitting.
3
Good Sharpe at 0 bps, collapses at 5 bps
The edge is in transaction timing, not signal quality. The strategy trades too frequently for costs to be sustainable. Try a lower-frequency strategy or reduce turnover.
4
Data error — ticker not found
The ticker is outside the reliable data contract. Try SPY first to confirm the strategy itself works. If SPY runs fine, swap your original ticker for a US equivalent.
5
Sharpe above 3.0 — suspiciously high
Almost certainly look-ahead bias, a data error, or overfitting to a very short in-sample period. Cross-check with /compare against simple buy-and-hold. Real strategies rarely exceed Sharpe 2.0 out-of-sample.
Demo — Step 1
Bad OOS Sharpe means fix the evidence, not the story
What is walk-forward validation?
Walk-forward validation splits your date range into sequential train/test windows and reports Sharpe across each test fold — a stronger overfitting check than a single OOS split.
1
Enable it under Validation Folds
In the backtest form, set Validation Folds to Sequential WF. This adds a second diagnostic layer on top of your main OOS split.
2
Start with 5 folds
5 folds gives you 5 independent test windows across your date range. Each fold trains on data up to a point and tests on the next window.
3
Check WF Avg Sharpe
If WF Avg Sharpe is much lower than IS Sharpe (gap > 0.5), the strategy is sensitive to time period. It learned a specific regime, not a general pattern.
4
Check Worst Path Sharpe
If the worst fold Sharpe is deeply negative, the strategy failed in at least one market regime. A robust strategy should have no deeply negative folds.
5
Use it as a second check, not a replacement
Walk-forward validation is additive on top of the OOS holdout split. Both are needed — the OOS split tests the end of the sample, WF tests consistency across all periods.
Demo — Step 1
Walk-forward folds test whether the edge survives time
How scoreboards work
Scoreboards rank paper strategies and portfolios by their forward (live) Sharpe — not by backtest results. Understanding what drives the ranking helps you use it honestly.
1
Only forward performance counts
Scoreboard ranking starts at the strategy's live start date — not at backtest creation. Any earlier history only initialises the signal; it does not seed performance.
2
Minimum live-day threshold is enforced
Strategies below the minimum threshold do not appear, regardless of how good their short-term return looks. This is intentional — short samples are dominated by noise.
3
Sort by forward Sharpe, not return
High return over a short live history is almost certainly luck. Forward Sharpe adjusts for volatility and gives a fairer view of consistency across different strategies.
4
Do not use the scoreboard to select strategies for backtesting
Picking the top-ranked strategy and running a backtest on it is a second-pass search. The OOS holdout cannot protect against selection bias introduced after the backtest was run.
5
Use it to monitor, not to discover
The scoreboard's real value is in tracking strategies you already selected on fundamentals. A strategy that degrades on the scoreboard is a retirement signal, not a re-tuning opportunity.
Demo — Step 1
Scoreboards rank forward paper evidence, not backtests
Why is my validation blocked?
A 'Blocked' validation status means the system found a reason to hold back the strategy before it can be promoted. This is a protective gate, not a failure — it tells you exactly what to fix.
1
Read the alert copy
Each blocked sleeve shows an alert below the status. It names the specific gate that triggered — benchmark fit, drawdown threshold, live-day minimum, or OOS Sharpe floor. Start there.
2
Benchmark fit alert
This fires when your strategy's tracking error is significantly wider than the best available alternative in the same asset class. It does not mean the strategy is bad — it means it diverges too far from a sensible reference. Check your benchmark choice and whether the strategy's universe matches it.
3
OOS Sharpe below floor
The out-of-sample Sharpe from the last qualifying backtest fell below the promotion threshold. Re-run the backtest with a longer date range and realistic costs. If OOS Sharpe is still negative, the strategy has not demonstrated edge.
4
Drawdown gate
Peak-to-trough drawdown exceeded the plan limit for the sleeve. This is often a position-sizing or leverage issue, not a signal issue. Reduce position size or tighten the stop rule in the strategy spec.
5
Resolve, then re-run the backtest
Most blocks clear once the underlying issue is fixed and a fresh backtest is promoted. Clearing the alert manually without fixing the root cause will not raise the sleeve's quality floor.
What does regime conditioning do?
Regime conditioning filters a strategy's signals through a market-state detector. When the regime is unfavourable, the strategy goes flat — reducing drawdown at the cost of fewer trades.
1
What regime detection does
A regime classifier reads macro signals — volatility, trend strength, breadth — and labels the current market as Risk-On, Risk-Off, or Neutral. Your strategy only enters positions when the regime matches its operating conditions.
2
When to enable it
Regime conditioning is most useful for trend-following and momentum strategies that suffer badly in choppy or mean-reverting markets. Mean-reversion strategies can benefit in the opposite direction.
3
Enable it in the backtest Options panel
Open the Options panel and set Regime to 'Strategy default'. The first run will show how much the filter adds or costs in Sharpe and trade count.
4
Check the trade count
Regime filtering significantly reduces trade count. If the strategy fires fewer than 20 trades in the OOS period, your Sharpe estimate becomes unreliable — the sample is too small.
5
Regime Switch Cost adds friction
Every regime transition incurs an extra cost (default 6 bps). This models real-world re-entry costs. If OOS Sharpe only improves with 0 switch cost, the benefit is not real.
How do I read alpha and beta?
Alpha is the return your strategy earns above what the benchmark would explain. Beta measures how much your strategy moves with the market. Both together tell you whether you have genuine edge or just market exposure.
1
Beta: how correlated you are to the market
Beta = 1.0 means your strategy moves in lockstep with the benchmark. Beta > 1 means more volatile than the market; Beta < 0 means it tends to move opposite. A long-only strategy will typically have Beta 0.3–0.9.
2
Alpha: return unexplained by Beta
Annualised Alpha is the residual return after removing the portion explained by market beta. Positive alpha means genuine edge; negative alpha means underperformance even after adjusting for risk taken.
3
A high return with Beta near 1.0 is not alpha
If your strategy has Beta 0.95 and beats the benchmark by 2%, most of that is just market exposure. Compare the same date range with Beta = 0 (zero-beta equivalent) to isolate true alpha.
4
OOS alpha matters — IS alpha does not
The alpha in the in-sample period is always inflated. Only the out-of-sample period alpha (last 20%) is an honest test. Negative OOS alpha is a retirement signal regardless of IS performance.
5
Use a relevant benchmark
Alpha is only meaningful relative to the right benchmark. For an equity momentum strategy, SPY is the right reference. For a bond strategy, use TLT or AGG. A poor benchmark choice makes alpha meaningless.
When should I retire a paper strategy?
Retiring a paper strategy is a quality decision, not an emotional one. There are specific signals that indicate a strategy has degraded — and specific traps that cause premature or overdue retirement.
1
Retire when forward Sharpe is persistently negative
A single bad month is noise. Three consecutive months of negative forward Sharpe — especially if drawdown is deepening — is signal. The strategy is behaving differently than backtested.
2
Retire when the regime has structurally shifted
If your strategy was built for trending markets and the market has been range-bound for 6+ months, continued underperformance is expected. Retire and replace with a mean-reversion strategy for the current regime.
3
Do not retire on a short drawdown alone
Every strategy has drawdown periods — even ones that will recover. Check: is drawdown within the historical max drawdown from the backtest? If yes, it is within the expected envelope and not a retirement signal.
4
Do not tune — retire cleanly
Adjusting parameters on a live sleeve invalidates its forward record. If the strategy is broken, retire it and backtest a new version from scratch. Do not patch a live sleeve.
5
Replace, do not leave a gap
After retiring, have a replacement ready. Leaving capital undeployed while searching introduces timing bias. Identify your replacement strategy before pulling the trigger on retirement.
Dashboard
Your research command centre
The dashboard gives you a live overview of your research activity, the market, and today's signals in a single view. Nothing here executes trades — it is a read-only summary panel.
Decision rules
Use the dashboard to pick a question, not to find answers. Everything here is context — the market snapshot, signals, and recent backtests are inputs to your research, not conclusions.
Click any ticker chip to run a pre-filled backtest. This is the fastest path from "I see SPY moving" to a testable hypothesis — it skips the form-filling step entirely.
Use Focus view when you want to study one asset in depth. The 3-Up and 6-Up layouts are for scanning across the watchlist; Focus gives you the full chart with hover inspection.
Strategy Library
Browse and select strategies to test
The library contains the built-in strategies plus any saved Strategy Studio strategies in your account. Public authoring is intentionally bounded here; the raw-code lab remains admin-only and is not part of the public product path.
Decision rules
Start with a Beginner strategy. 1–2 parameters means fewer ways to overfit. Understand the signal logic before adding complexity — advanced strategies are harder to attribute, not inherently better.
Test across categories before committing to one. Momentum and mean-reversion strategies often take opposite sides of the same market move. A result that only works in one category may be regime-specific.
Read the signal logic before running the backtest. The param chips on each card show the tunable parameters. If you don't understand what the strategy is measuring, you can't interpret its results honestly.
| Category | What it bets on | Best for |
|---|---|---|
| Mean Reversion | Price reverts to a mean; trade the deviation back. | Range-bound markets, low-momentum regimes |
| Momentum | Trend-following; buy strength, sell weakness. | Trending markets, strong macro themes |
| Volatility | Trade volatility expansions or contractions. | Event-driven trades, options-aware sizing |
| Factor | Systematic factor exposures: value, carry, size. | Multi-asset or long-horizon allocations |
| Statistical | Model-driven frameworks: Kalman, OU, cointegration. | Pairs trading, spread strategies |
Backtest
Run a single strategy on historical data
The backtester runs a chosen strategy on a chosen ticker over a date range and returns an equity curve plus a full metrics table. All results are stored in your account and accessible from the Dashboard.
Decision rules
Always set realistic costs before judging any result. Start at 5 bps commission + 3 bps slippage. A strategy that only works at 0 bps has no practical edge — transaction costs will erase it live.
Judge by OOS Sharpe, not IS Sharpe. IS Sharpe is almost always inflated. A strategy is credible when OOS Sharpe ≥ 0.5 and the IS/OOS gap is ≤ 0.5. A negative OOS Sharpe is a failure — do not proceed.
Use at least 5 years of data. Shorter ranges have too few independent market regimes to say anything meaningful. For mean-reversion strategies, 7+ years is safer.
Ask: can I hold through 2× the backtested max drawdown? If the answer is no, the strategy is already too volatile for you — find one with a smaller drawdown rather than telling yourself you'll hold.
| Parameter | What it controls | Recommended start |
|---|---|---|
| Commission | One-way transaction cost per turnover dollar (basis points). | 5 bps — typical retail broker |
| Slippage | Execution friction on top of commission; models bid-ask spread. | 3 bps liquid ETFs; 8–15 bps small-caps |
| OOS Split % | Reserves the last N% of the date range as an untouched holdout. | 20% — lower only if date range ≥ 10 years |
| Validation folds | Walk-forward or combinatorial CV diagnostics on top of the main run. | Sequential WF 5-fold if date range ≥ 4 years |
| Benchmark | Buy-and-hold reference for equity curve comparison. | SPY for equity strategies; same ticker for others |
Free-Plan Core Tickers
The practical safe set when validation fails on Free
This is the practical core universe for the Free tier, not an exhaustive entitlement promise. Stay inside these names when you want the clearest supported path. Lookup is discovery only. Resolve is the actual support-contract check for the requested window.
Decision rules
Use US ETFs for your first backtests. SPY, QQQ, IWM, GLD, TLT have the cleanest daily data, the longest history, and the most reliable validation on the Free plan.
Non-US tickers with country suffixes (.L, .DE, .T) require Pro. If you're on Free, stay inside the core US universe.
If a backtest fails with a data error, try SPY first. If SPY runs fine, the original ticker is outside the reliable contract. Swap for a US equivalent and note which tickers caused issues.
Core ETFs
SPY, QQQ, IWM, DIA, VTI, EFA, EEM, XLF, XLE, XLK, XLV, ARKK
Core single names
AAPL, MSFT, NVDA, GOOGL, AMZN, TSLA, META, JPM
Rates and real assets
TLT, IEF, HYG, AGG, GLD, SLV, USO
Core crypto
BTC-USD, ETH-USD, SOL-USD
Major indices
^GSPC, ^SPX, ^NDX, ^IXIC, ^DJI, ^RUT, ^VIX
Rule of thumb: If you want non-US suffixes, niche indices, or weighted benchmark baskets, you are outside the narrow Free reliability contract and into Pro best-effort territory. Unsupported names are outside both plan promises until telemetry justifies review and the contract is updated deliberately.
Metric Reference
What each number in a backtest result means
Decision rules
Metric reference
IS Sharpe is an upper bound, not a target. Always above zero for a fitted strategy. Values > 1.0 are acceptable; > 1.5 are strong — but these numbers are inflated by in-sample fitting and should be treated with scepticism.
OOS Sharpe gap thresholds: < 0.3 = clean (low overfitting); 0.3–0.5 = acceptable; > 0.5 = red flag; negative OOS Sharpe = the strategy failed — do not promote.
Win rate below 50% is fine — if payoff ratio is above 2:1. A strategy can be highly profitable with a 40% win rate if its average winner is 3× the average loser. Check both together, not in isolation.
Max drawdown test: Can you psychologically hold through 2× this drawdown? If no, the strategy is already too volatile. Do not convince yourself you'll behave differently under real losses.
Overfitting warning: IS Sharpe is almost always better than OOS Sharpe. A gap > 0.5 or a negative OOS Sharpe strongly suggests the strategy is curve-fitted to the training data and should not be traded.
Strategy Comparison
Rank multiple strategies side-by-side
The comparison page runs several strategies on the same ticker over the same period and overlays their equity curves. It is the fastest way to see which strategies actually differ versus which ones are only telling different stories about the same underlying exposure.
Decision rules
Select at least one strategy from each category. One momentum, one mean reversion, one volatility. If yours is the only one with a good OOS Sharpe across all three, it may be the lucky draw from a large search, not genuine edge.
If two equity curves converge tightly, they are trading the same bet. No diversification benefit from using both. Genuine alternatives diverge — especially during regime transitions like 2020, 2022.
Sort by OOS Sharpe, not CAGR. High CAGR often means more volatility or leverage, not more skill. OOS Sharpe adjusts for both and is the only metric that captures holdout performance.
Tip: Select 4–6 strategies from different categories (one momentum, one mean reversion, one volatility) to understand whether your best backtest result is genuine or just the lucky draw from a large strategy set.
Paper Trading
Track strategies forward from today
Paper trading creates a forward record from the first live day after creation and tracks hypothetical P&L from that point onward. No capital is deployed — this is a forward-monitoring layer, not a broker bridge.
Decision rules
Promote to paper when: IS Sharpe ≥ 1.0 and OOS Sharpe ≥ 0.5. The IS/OOS gap should be ≤ 0.5. If OOS Sharpe is negative, the strategy is likely overfit — do not paper it.
Choose long-only unless short signals were tested and profitable. Long/short sleeves amplify both gains and losses. Most built-in strategies are long-biased by design — adding short exposure without testing it is uncontrolled risk.
Wait 60 live days before drawing any conclusions. Scoreboard rankings require a minimum threshold. Anything under 30 days is dominated by noise — a strong first month is not evidence of edge.
Retire when: forward Sharpe drops below 0.3 for 30+ consecutive days, or live max drawdown exceeds 1.5× the backtested drawdown. These signal a regime shift — not a tuning opportunity.
| Parameter | What it controls | Recommended start |
|---|---|---|
| Strategy | The rule set generating LONG / SHORT / FLAT signals. | Same strategy as your OOS-validated backtest |
| Ticker | Asset tracked; fills at daily close, no slippage model. | Same ticker as your backtest — don't switch asset class |
| Sleeve type | Long-only mirrors LONG signals; Long/short also goes short. | Long-only unless short signals were tested and positive |
| Refresh | Updates live NAV from latest close. Auto at 18:00 UTC. | Let auto-refresh handle it; manual only for same-day check |
| Portfolio weight | Share of portfolio NAV attributed to this sleeve. | Equal-weight to start; adjust after 60+ live days |
Note: Paper fills happen at daily close prices with no slippage model. This is forward simulation, not live trading. Real fills will differ — paper performance measures signal character forward, not replicable trade P&L.
Signal Monitor
Today's strategy signals across the watchlist
The signal monitor runs all built-in strategies across a curated watchlist of tickers and reports the current signal direction and strength for each combination. Signals are computed from daily bars — they update once per day, not intraday.
Decision rules
Filter to LONG or SHORT with min strength ≥ 0.5 before acting on any signal. FLAT is the default state — most ticker/strategy pairs are FLAT most of the time. Unfiltered, the monitor is noise.
Always backtest the specific strategy on the specific ticker before treating a signal as actionable. Strength measures persistence, not quality. A persistent signal from a bad strategy is still bad.
Signals are computed from daily bars — they update once at 18:00 UTC, not intraday. A signal on day t reflects the strategy logic through close of day t. Do not act on intraday price moves expecting signal changes.
| Signal | What it means | When to act |
|---|---|---|
| LONG | Strategy logic indicates a long position as of the last daily close. | After verifying backtest — OOS Sharpe ≥ 0.5 |
| SHORT | Strategy indicates short, or absence of long on long-only strategies. | Only if strategy was tested with shorts; verify backtest |
| FLAT | Strategy is neutral — no position recommended. | Normal state — do not act; use filters to hide |
| Strength | Normalised [0,1] persistence and magnitude of signal. | Use ≥ 0.5 as minimum; ≥ 0.7 is high-conviction |
Scoreboards
Forward performance rankings
Scoreboards rank paper strategies and portfolios by their forward (live) performance — not by historical backtest results. The key distinction is that stored live history starts only after creation, while any earlier lookback is used only to initialize signal state.
Decision rules
The minimum live-day threshold is intentional, not a bug. A strategy with 5 days of live history appearing at the top of a leaderboard is meaningless. The threshold forces genuine forward exposure before ranking begins.
Sort by forward Sharpe, not by return. High return over a short live history is almost certainly luck. Forward Sharpe adjusts for volatility and gives a fairer view of consistency.
Do not use scoreboard position to select strategies for backtest. If you backtest every top-ranked strategy looking for one with good IS metrics, you are doing a second-pass search that the OOS holdout cannot protect against.
Important: A raw backtest leaderboard would mostly reward overfitting. This platform deliberately does not rank strategies by in-sample backtest returns.
Methodology
How the engine works — implementation details and limits
Key implementation limits to keep in mind
All signals use next-bar execution. A signal on day t is applied to the return from t → t+1. There is no intraday fill model, no open-vs-close microstructure, and no execution latency. Real execution will differ from the modelled return.
The cost model is conservative but incomplete. Commission and slippage are deducted per-turnover. There is no market-impact model, no partial-fill simulation, no borrow-fee model for shorts, and no venue-specific open execution model. For large sizes or illiquid assets, real costs will be higher.
The OOS holdout is diagnostic evidence, not proof of robustness. It is not a nested train-tune-test optimisation workflow. Use it to falsify — a negative OOS Sharpe is a clear failure signal. A positive OOS Sharpe is encouraging, not conclusive.
Signal timing
Signals are computed from daily OHLCV bars. A signal on day t reflects the strategy logic applied through close of day t, and the backtest applies that position to the next bar return t→t+1.
No execution latency model, no open-vs-close microstructure model, and no intraday fill guarantee.
Backtest return model
Position on bar t is applied to the forward return r(t→t+1). Commission and slippage in bps are deducted via realised turnover, and regime-conditioned overlays can add a dedicated switch-friction penalty on confirmed state changes and de-risking ramps.
Still no full market-impact model, partial-fill simulation, borrow-fee model for shorts, or venue-specific open execution model.
Out-of-sample split
The last N% of the date range is withheld as a terminal holdout. OOS metrics are shown separately, while the headline Sharpe/CAGR/MaxDD remain full-sample metrics for the full run.
Still not a nested train-tune-test optimisation workflow. The holdout is diagnostic evidence, not a proof of robustness.
Walk-forward validation
You can run either sequential walk-forward folds with a purge gap or combinatorial purged CV. Both report fold/path count, average Sharpe, worst-path Sharpe, and positive-path ratio, and both are additive diagnostics on top of the main run.
This is stronger than a single holdout, but still not proof. There is no nested parameter search, no execution simulator, and no full institutional research stack.
Benchmark comparison
Default benchmark is the same-ticker buy-and-hold. You can override with a single ticker or a weighted basket, and result surfaces now also compute constructed alternatives such as a factor-matched optimizer basket and a risk-balanced core benchmark.
Constructed benchmarks are still reference tools, not a claim that the optimizer found the “true” neutral benchmark for the strategy.
Paper scoreboards
Scoreboards rank forward-tracked paper strategies and portfolios from stored live history only. Warmup history can initialize state, but live NAV is reset at the forward start and minimum live-day thresholds still apply.
Still not a brokerage-linked live track record, and the board remains a product ranking surface rather than an institutional performance composite.
Risk Disclosure
What this tool does not and cannot tell you
Not financial advice. Alphrex is a research and simulation tool. Nothing on this platform constitutes investment advice, a trading recommendation, or a solicitation to buy or sell any financial instrument.
Past backtest performance does not guarantee future results. Backtests are model outputs conditional on data quality, cost assumptions, look-ahead-bias absence, and implementation details. A strategy that performed well in backtest may fail to survive:
Key risks to understand before acting
Regime shifts: A momentum strategy that worked in 2010–2020 trending markets may fail in a choppy or mean-reverting regime. Past backtest performance in one regime is not a guarantee of future performance in another.
Crowding: When a strategy becomes widely followed, its edge erodes as more capital competes for the same signal. A backtest conducted on historical data cannot detect current crowding levels.
Live execution frictions: Real fills differ from modelled fills. Slippage estimates are conservative approximations — illiquid names, large orders, and volatile opens will incur larger real-world costs.
Survivorship bias: Data is biased toward companies that survived. Backtesting on today's S&P 500 constituent list implicitly selects for winners and overstates returns for stock-selection strategies.