Stock Backtest Assumptions and Limitations
Understand Tessera’s historical portfolio simulations: dated factor inputs, portfolio rules, price and dividend conventions, costs, and known methodological limits.
A Tessera backtest is a historical simulation of portfolio rules. It is not an actual investment account, an achieved investor return or a forecast. This page explains the implementation assumptions and limits that matter when interpreting a run.
Public research pages do not present backtest results as product performance claims. Historical simulations, paper portfolios and the track record are different reporting contexts; inspect the methodology and dates for the surface you are using.
What is being simulated?
The portfolio engine selects a dated factor-panel slice, scores eligible stocks under the chosen preset, applies decision rules and constructs a target portfolio. It then compares the target with simulated holdings and executes the resulting orders using historical prices.
The scoring and portfolio construction code is shared across historical and ongoing modes. Execution, data availability and some risk checks differ. Shared code improves consistency of the rules; it does not establish identical real-world outcomes.
A run depends on more than a preset name. Its dates, rebalance schedule, portfolio policy, permitted overrides and data version all shape what is tested. The factor guide and portfolio construction guide describe those rules.
Historical data and coverage
The engine reads a panel slice on or before the requested decision date, within its supported lookup window. The selected panel date can differ from the calendar rebalance date. A requested start and end do not prove that every input has complete coverage throughout that interval.
We do not promise a universal ten-year point-in-time history, complete delisted-stock coverage or original-publication values for every provider field. Symbol changes, acquisitions, missing prices, source revisions and current classification metadata can affect results.
Read survivorship, look-ahead and overfitting for why these limitations are separate problems.
Simulated fills and trading costs
The standard rebalance broker uses historical opening prices with a fixed slippage adjustment, currently 10 basis points per side. This is a model assumption, not a measurement of your expected execution costs. Historical-price availability and the surrounding workflow can affect the prices supplied to that broker.
A fixed adjustment does not reconstruct bid-ask spreads, market impact, partial fills, order-book depth or broker outages. Other workflow paths can use different fill conventions. Do not assume every simulated event is a next-session-open order with identical cost treatment.
Returns also do not represent your personal tax treatment, subscription cost or every charge a broker may levy. A modeled cost cannot guarantee realistic execution for a particular order size or security.
Dividend treatment
Available dividend events are credited as cash to held positions using the workflow’s ex-date convention. Cash may later fund portfolio orders; the simulation does not automatically reinvest every dividend into the same stock.
Data completeness and consistent benchmark treatment matter. Read price return versus total return before comparing series with different dividend conventions.
Risk rules do not all operate identically
The daily disaster-stop check in paper and live operation is not applied by the backtest engine. Separately configured entry-stop rules and monitoring policies can have their own behavior. A historical result must not be interpreted as evidence that every ongoing safeguard was simulated.
Target weights, sector limits and rebalancing bands constrain planned construction. They do not guarantee a maximum realized loss or an exact invested weight. The position sizing guide explains that distinction.
Hindsight and selection risk
A simulation can faithfully reproduce a rule that was selected with hindsight. Trying many presets, weights, filters or windows and keeping the most favorable outcome introduces selection risk. No user backtest automatically becomes a valid out-of-sample experiment.
Record the configuration before comparing periods. Keep an untouched evaluation window where possible, inspect sensitivity to reasonable changes and account for how many alternatives you tried. These practices improve research discipline without making future performance predictable.
Read how paper portfolios work to distinguish ongoing simulations from historical tests. For current access to backtests and paper portfolios, see pricing. Use the methodology index to connect the simulation with the underlying scoring and portfolio rules.