How we evaluate
investment performance.

Our public performance page presents historical portfolio simulations built from batch rankings. It shows what specified investment rules produced under the stated data and execution assumptions. It is not the performance of a live fund, a client account or an independently audited investment record.

A credible comparison needs a defined period, consistent valuation, an appropriate reference and clear assumptions. The sections below connect those reporting principles to the actual calculations on this site.

In these simulations, all capital is supplied at the start. The cumulative return is therefore (ending wealth ÷ starting wealth − 1) × 100. Purchases, sales, rebalances and dividends move value within the portfolio; they are not new investor contributions.

If deposits or withdrawals were introduced, that simple formula would no longer isolate investment performance. A time-weighted return links returns between external cash flows; a money-weighted return also reflects the amounts and timing of those flows. These answer different questions. Our current fixed-capital simulation has no later external cash flows.

Wealth includes the cash wallet, long holdings, the liability to close short holdings, and net unpaid dividend receivables. Cash Balance separately shows available cash after reserved short collateral. Uninvested cash remains part of the return calculation even when it earns nothing.

Dividend entitlement is recognized at the modeled ex-dividend date. On the payment date, the receivable becomes cash: payment must not count the same income a second time. Paid dividends wait until the next batch for strategy purchases. Unknown payment dates leave an unpaid receivable, so that amount contributes to modeled wealth but cannot fund a purchase.

All values are expressed in USD. Security prices and dividend payments can use different currencies or quote units; converting a dividend requires its own currency, and a GBX price requires conversion from pence before applying a GBP exchange rate. Currency movements can therefore affect the USD result independently of the local share price.

Displayed returns cover the actual selected dates. We do not turn a partial year into an achieved annual return. The interest comparison is different: its annual rate is an explicit assumption, compounded over elapsed calendar days. It is not evidence that a strategy will keep earning at its recent pace.

SPY provides an S&P 500 ETF reference; QQQ provides a Nasdaq-100 ETF reference. Each starts with the same $100,000 on the selected starting date and is held through the same ending date. Net dividend payments stay in cash. These modeled ETF portfolios are not the published dividend-reinvesting total-return indices.

Both references provide useful market context, but neither matches every portfolio here. Our strategies may hold international stocks, fewer companies, substantial cash or short positions. QQQ also has a different sector mix from SPY. A higher return than either ETF does not, by itself, establish superior risk-adjusted skill.

When comparing cumulative returns, a strategy returning 12% against a benchmark returning 8% leads by 4 percentage points. That simple difference is not statistical alpha. Establishing alpha requires a specified risk model and enough evidence to assess uncertainty.

The editable interest line offers a separate opportunity-cost reference. Its default 3.5% annual rate is hypothetical, not a current Treasury yield, a guaranteed deposit offer or a risk-free rate observed throughout the historical period.

Maximum drawdown describes a loss experienced along the path, even if the portfolio later recovered. A 20% fall requires a 25% subsequent gain to return to the starting value. The worst observed drawdown is not a limit on future losses, and daily values do not capture every intraday move.

Inspect concentration, cash exposure and short positions alongside these figures. Short-sale proceeds are accompanied by a liability and collateral requirements; they are not free profit. Short positions can lose more than their initial sale value, and a simulation without borrow availability or financing costs cannot establish real-world executability.

No single risk-adjusted score settles the question. For example, a Sharpe ratio relates average excess return to return variability; it does not directly describe tail losses, liquidity or execution constraints. Any such statistic needs its return frequency, reference rate and observation window specified. A short, overlapping history cannot establish persistent skill.

A forecast can get the direction right and the magnitude wrong. It can identify a profitable company while underperforming the market. It can anticipate the eventual move but arrive too early for the chosen investment horizon. These are different outcomes, so one percentage cannot represent all of them.

Direction accuracy asks whether price moved up or down as predicted over a defined window. Return error asks how far the outcome was from the forecast. Neither, alone, tells you what a portfolio earned or what losses an investor had to endure.

The forecast and outcome must use the same definition. A price-return forecast should be checked against price return; a forecast including dividends should be checked against a corresponding income-inclusive outcome. Tax treatment, currency and dates must be stated. Comparing one basis with another can create apparent error that comes from the measurement itself.

The horizon also matters. A one-year forecast observed for three months has not reached its evaluation date. Comparing its interim return with a proportionally scaled target can describe progress under an explicit assumption; it does not prove the forecast correct or incorrect. Share returns need not arrive at a constant monthly pace.

Equally, a low hit rate is not evidence of skill just because a few large gains are possible. Direction, forecast error, ranking quality and portfolio outcomes should each be measured on a defined population and horizon. None should be substituted for another when a result disappoints.

An illustrative 50% direction hit rate across a batch of 400 assets describes the broad research universe. It does not establish whether the small set of strongest ranked opportunities performed well. Those populations must be evaluated separately.

iPulse AI Top Picks helps surface stronger buy signals and agreement across advisor forecasts. Our portfolio research asks what happened when explicit rules selected and allocated capital to a small ranked set. The score incorporates more than agreement alone; see the consensus scoring methodology.

A selected portfolio can outperform even when broad-universe accuracy is modest, but that must be demonstrated rather than assumed. A 50% hit rate is not automatically the right baseline in a market where most assets rose. Repeated forecasts for the same assets and overlapping batches are also correlated observations, not wholly independent trials.

The current historical portfolios use original quantitative batch rankings. They do not backtest the separate editorial Top Picks process that was added later. The two are linked research surfaces, but they are not interchangeable historical records.

The Performance & Strategies page applies stated rules to an initial $100,000. Choosing another starting batch creates a fresh investment from that batch; it does not crop an already compounded portfolio.

Read total return alongside excess return versus benchmarks, maximum peak-to-trough drawdown, concentration and cash exposure. Forecast direction and error remain useful diagnostics. A portfolio simulation brings those forecasts together with allocation decisions; it is not a substitute for honest out-of-sample evaluation.

A ranking is an input, not a complete trading strategy. You still need to decide how many assets to hold, how much to allocate, when to exit, how to rebalance and what to do with dividends.

• Selection: five, ten or twenty buys; a long-only portfolio or one that also shorts low-ranked assets.
• Position size: equal allocations or more capital for higher ranks. The displayed strategies apply a 35% single-underlying-asset cap at modeled daily checks.
• Exit rules: rebalance at the next batch, sell at a 25% price gain, or exit at 130% of the original one-year projected upside. A 20% forecast upside therefore gives a 26% gain target.
• Dividends: spend cash immediately after payment, reinvest into the paying asset, or wait for the next batch. These are different strategies. In the displayed simulations, paid dividends and exit proceeds wait for the next batch; the ETF benchmarks hold dividends as cash.

We display five understandable rules from a broader strategy research set. They are selected examples, not proof of an optimal portfolio, personalized recommendations, or a promise that they suit the average investor. The featured set was chosen after historical results were available; that selection itself can favor past winners.

Batch inputs and models evolve. Global context first appears in Batch 3; equity fundamentals and asset-consensus synthesis in Batch 5; Batch 6 includes currency-normalized fundamentals. The expandable comparison below the cash chart shows the models, actual forecast-input ranges and critical capabilities for each batch, newest first.

Features added after a run are labeled as such. An improvement to the process is not proof of improved forecasting performance. Later batches have shorter observation windows and different market conditions, so both per-batch and multi-batch results matter. Longer history across different market regimes makes evaluation more informative; short histories cannot establish stable skill.

Explore the engine history and lessons learned for the underlying changes and their limits.

The current results are retrospective forecast-date reconstructions. Some ranking snapshots were archived after those dates, so these curves do not establish that the exact portfolios were tradable with the same information at the time. A recorded-availability evaluation answers a different question and must be labeled separately.

Prices, currencies, splits, dividend entitlements, payment dates and modeled withholding affect the results. Separately modeled trading commissions, slippage, financing and short-borrow costs are excluded. ETF operating expenses already reflected in ETF prices are not added back. “After dividend withholding” therefore does not mean after every investor tax or investment expense.

Testing many rules and highlighting the strongest can favor chance winners. Stronger evidence comes from fixing selection and trading rules before a new evaluation period, recording when inputs became available, and retaining unsuccessful results as well as successful ones. Historical universes must also account for assets that disappeared or became untradable; a current list of surviving companies is not sufficient to establish freedom from survivorship bias.

We publish the valuation date and preserve complete data generations. A failed daily refresh must not overwrite a complete history. Reproducibility makes a calculation inspectable; it does not remove hindsight, missing data or execution assumptions. Historical performance does not establish future returns.