2026 Grid Bot Exit Guide
When to Stop a Grid Bot: 9 Exit Signals & Decision Rules (2026)
A grid can keep printing small completed trades while the strategy as a whole is losing money. This guide replaces emotion with a repeatable framework for choosing whether to wait, pause, recenter, or close.
Quick answer
Stop a grid bot when the reason you launched it is no longer true—not merely because today is red or because the bot shows a positive grid-profit figure. The strongest exit signals are a confirmed regime change, a breached risk budget, persistent price displacement outside the range, net grid edge below costs, deteriorating liquidity, a reached strategy target, an operational failure, excessive opportunity cost, or unacceptable futures liquidation risk. Before closing, compare total equity with the starting value and confirm what the platform will do with open orders and remaining inventory.
Affiliate disclosure: TradeboticsAI may earn a commission if you join a platform through a link on this page, at no additional cost to you. That does not change the analysis, risk warnings, or decision rules below.
Bitsgap includes demo trading, bot controls, and performance tracking so you can rehearse a shutdown workflow first.
Try Bitsgap in Demo Mode
The core rule: stop the thesis, not the screen color
The right answer to when to stop a grid bot starts before the bot launches. A grid is a market hypothesis encoded as orders: price should oscillate often enough inside a defined range for completed buy-low/sell-high cycles to exceed fees, spread, slippage, and the loss or opportunity cost attached to inventory. If you never wrote that hypothesis down, every drawdown feels like a reason to quit and every green day feels like permission to continue.
Create an exit plan with four parts: the market condition that must remain true, the maximum total-equity loss you accept, the evidence that counts as a genuine range break, and the exact action taken when a trigger fires. A statement such as “run while the daily market remains range-bound between the established support and resistance; review after two daily closes outside the range; close if total bot equity falls 8% from start” is useful. “Stop if it looks bad” is not. In practice, when to stop a grid bot becomes much easier to answer once every trigger is observable and written down.
This distinction matters because a grid bot has no independent judgment. It repeats the instructions you supplied. Automation removes manual order entry; it does not remove market risk or repair a stale premise. BloFin’s 2026 grid guide similarly frames a stop around changed structural conditions—such as a trend breakout, collapsed volatility, or weaker liquidity—rather than around impatience alone. See the BloFin grid trading guide.
Thesis
What must stay true?
Define the range, expected oscillation, time horizon, pair quality, and reason the setup has positive net expectancy.
Invalidation
What proves it wrong?
Use closes, volatility, equity loss, liquidity, or operational evidence—not a single frightening candle.
Action
What happens next?
Preselect wait, pause, recenter, or close, including the treatment of leftover base assets and derivatives positions.
First, separate grid profit from total P&L
The most dangerous dashboard misunderstanding is treating “grid profit” as the final result. Grid profit usually counts completed buy-and-sell cycles. It can be positive while the unsold base asset has fallen enough to make the bot’s total result negative. Conversely, a bot can show modest grid profit while an appreciating inventory position makes total equity positive. The decision should therefore use a marked-to-market view of everything the bot owns and owes.
This is an analytical approximation, not a substitute for the platform’s ledger. Different providers label components differently. Pionex distinguishes grid profit, released profit, a take-profit closing trigger, and total profit; its current explanation warns that a take-profit label is a closing rule, not a guaranteed profitable outcome. Review Pionex’s take-profit explanation and its grid-profit versus release-profit guide for the platform-specific terminology.
| Metric | What it tells you | What it can hide | Use in an exit decision |
|---|---|---|---|
| Grid profit | Result from completed grid cycles | Unrealized loss on accumulated base asset | Evidence of activity, not sufficient evidence of success |
| Unrealized P&L | Current mark-to-market inventory change | Future execution price and exit slippage | Essential for a total-equity view |
| Total profit / bot value | Broader strategy result, depending on provider | Definitions, delayed marks, funding, or external transfers | Reconcile with the exchange balance and ledger |
| Profit per grid | Estimated or realized edge for one interval | Trade frequency and portfolio drawdown | Compare with round-trip costs |
| Available balance | Funds not locked in current orders | Assets reserved, positions, and liabilities elsewhere | Useful only after full account reconciliation |
For a deeper treatment of cost drag, read our crypto grid bot fees guide. If you need to estimate the relationship between spacing, fills, and costs before launch, use the crypto grid bot profit calculator.
9 signals that tell you when to stop a grid bot
No single indicator works for every pair or time frame. Use a hierarchy: hard risk triggers override everything; structural invalidation comes next; efficiency and opportunity-cost signals prompt review rather than instant liquidation. The nine signals below deliberately separate “the strategy is unsafe” from “the strategy is merely quiet.” That hierarchy is the most reliable way to determine when to stop a grid bot without reacting to ordinary noise.
1. The range thesis is structurally invalidated
A wick outside the boundary is not automatically a regime change. Crypto markets frequently sweep visible levels before returning. Stronger evidence is a close beyond the range on the time frame used to design the bot, followed by continued acceptance outside it: repeated closes, expanding volume, a retest that holds, or a sequence of higher highs and higher lows above resistance (the reverse below support). Your predetermined rule might require two daily closes, one weekly close, or a volatility-adjusted distance. Pick it before the breach.
If price has established a directional regime, a stationary grid is solving the wrong problem. Above the upper boundary, a spot grid may have sold most base inventory and become idle in quote currency. Below the lower boundary, it may have accumulated the asset and stopped placing lower buys. Neither state is necessarily catastrophic, but the original oscillation engine is no longer operating as intended. Consider closing, or start a separately evaluated grid only after the new range is mature; do not shift boundaries simply to avoid recognizing a loss.
2. Total equity reaches the predefined risk budget
A hard risk limit is the clearest answer because it converts an open-ended judgment into a bounded decision. Measure from a consistent starting equity and specify whether the threshold uses mark price, last price, or a confirmed account value. The trigger should include realized grid gains and unrealized inventory losses. For leveraged bots, include funding and any additional margin transferred after launch; otherwise the apparent percentage can understate the capital committed.
A breached risk budget should normally cause closure or an immediate reduction, not a debate about whether the asset “must bounce.” Adding capital after the threshold changes the strategy and can enlarge the loss. If your tolerance is 6%, 8%, or another value, that is a personal risk choice—not a universal recommendation. The important part is sizing the bot so the threshold is financially and emotionally survivable.
3. Price remains outside the range long enough to make the bot idle
Out-of-range is a state, not an automatic verdict. A spot grid can resume if price returns, and waiting may be reasonable when the breakout is small, the broader range remains intact, and capital has no better planned use. But persistent displacement creates inventory concentration and opportunity cost. Track time outside the grid relative to your original trade horizon, distance beyond the boundary in volatility units, and whether the bot still has orders capable of filling.
Gainium’s documentation notes that a bot out of range will not perform normal two-sided grid trading until price returns to the configured area. Read what happens when price is out of range. A practical review trigger might be “three daily closes outside, more than one average daily range beyond the boundary, and no grid cycle for five days.” The numbers must match your own system; the purpose is to prevent indefinite waiting by default.
4. Net profit per completed grid no longer clears costs
A busy bot can destroy value slowly. Estimate the completed grid’s gross percentage, then subtract both sides of trading fees, expected spread, slippage, and—where applicable—funding or borrowing. If the remaining margin is negligible or negative, more fills are not a cure. Narrow grids are particularly vulnerable because costs consume a larger share of each cycle.
Review actual fills rather than the setup screen alone. Maker orders can execute as taker orders in some circumstances, fee tiers can change, low-liquidity pairs can widen abruptly, and a market exit adds another cost. If the bot only works under an optimistic fee assumption, pause it and redesign the spacing. Our grid bot settings guide explains how range width and grid count interact with net edge.
5. Volatility collapses or changes shape
A grid needs movement through multiple levels, not volatility for its own sake. When realized movement contracts below the distance required to complete cycles, the bot may leave capital idle while subscription fees or opportunity costs continue. At the other extreme, one-way volatility can race through the entire range and concentrate inventory. Both are mismatches even though one market looks quiet and the other looks exciting.
Compare recent average true range, completed cycles per day, and the distribution of moves with the period used to select the parameters. A gradual decline in fills usually calls for review, pause, or a new backtest—not an emergency market sale. A sudden volatility expansion paired with a range break deserves faster action. Historical testing can reveal how often the setup became inactive; see our crypto grid bot backtesting guide.
6. Liquidity and execution quality deteriorate
The visible market can change while price remains inside the lines. Warning signs include a wider bid-ask spread, thinner depth near each grid level, repeated partial fills, larger slippage, exchange-specific price gaps, or sharply lower volume. These conditions reduce realized edge and make an eventual exit more expensive. They are especially relevant for small-cap pairs, newly listed tokens, delisting risk, or a venue with impaired withdrawals.
Do not rely on headline volume alone. Inspect the order book near your typical order size and compare the bot’s actual execution with the mid-price. If liquidity has migrated to another venue, moving the bot requires a fresh counterparty and operational review; it is not just a parameter edit. A pair that is technically still trading may no longer be suitable for automation.
7. The target is reached or the remaining reward is no longer worth the risk
A profitable bot also needs a stopping condition. Targets can be based on total equity, time, a market price, or a risk-adjusted objective. The cleanest target specifies the metric and action: “close when marked bot equity is 12% above starting equity and convert the remaining base asset to quote currency,” for example. A vague desire to “let winners run” can turn a completed plan into an unmanaged exposure.
Platform controls vary. Gainium documents price-based and bot-value percentage triggers, with actions that can either cancel orders only or cancel and sell the base asset. See its grid take-profit and stop-loss documentation. Confirm the live platform behavior because product settings can change and a trigger price does not guarantee an identical execution price.
8. Operational integrity fails
Stop or isolate the strategy when the automation can no longer be trusted. Examples include revoked API permissions, persistent synchronization errors, duplicate or missing orders, exchange maintenance during a critical move, an unexplained balance mismatch, stale price data, failed stop orders, or a provider incident. “The bot says active” is not proof that protective orders exist on the exchange.
Use read-only reconciliation where possible: compare provider records with exchange open orders, trade history, balances, and derivatives positions. If an API connection is unstable, avoid frantic repeated commands that could create duplicates. Capture evidence, prevent new exposure if the platform allows it, and follow the venue’s official incident instructions. Security or custody concerns outrank a hoped-for grid return.
9. Futures margin or liquidation risk becomes unacceptable
For a futures grid, waiting outside the range is not equivalent to waiting with unleveraged spot inventory. Directional exposure, maintenance margin, funding, and liquidation distance can change quickly. A neutral grid can acquire net exposure as fills accumulate. Cross margin can also place more of the account at risk than the bot’s displayed allocation suggests.
Close or reduce when liquidation distance violates your minimum buffer, margin usage exceeds the plan, funding materially changes the edge, or the exchange position cannot be reconciled. Do not treat an extra margin transfer as proof that the thesis improved. If you are still deciding which structure fits your risk, compare spot grid versus futures grid before deploying leverage.
Wait, pause, recenter, or close?
Knowing when to stop a grid bot is not binary. “Do nothing” and “sell everything” are only two of four useful actions. The correct choice depends on whether the thesis survives, whether exposure remains acceptable, and whether the platform can preserve the current state safely.
| Action | Use it when | Do not use it to | Required check |
|---|---|---|---|
| Wait | Thesis remains valid, risk budget is intact, and displacement is temporary | Avoid admitting that a confirmed trend invalidated the range | Define the next time/price review and capital opportunity cost |
| Pause | Execution or event risk is temporary and the platform can halt new orders safely | Assume existing inventory or futures exposure disappears | Verify what “pause” does to open orders and positions |
| Recenter | A new range is established and a fresh test supports new parameters | Move the goalposts around a losing position | Treat it as a new strategy with new baseline and risk limit |
| Close | Hard risk trigger, thesis invalidation, target, integrity failure, or unsafe leverage occurs | React blindly to one ordinary candle | Confirm order cancellation, inventory handling, and final balance |
Wait
Waiting is an active decision only when it includes a deadline and invalidation level. Record why you expect re-entry, how long you will tolerate inactivity, and what the capital could earn or protect elsewhere. A bot sitting above its upper range in quote currency may be relatively simple to monitor; a bot below the lower range holding a volatile base asset has a different downside profile.
Pause
Pause can be appropriate ahead of planned exchange maintenance, a known high-impact event, or a recoverable API problem. But platform definitions differ. Some pause functions stop new logic while leaving exchange orders live; others cancel orders but retain inventory. Verify rather than infer. In derivatives, a paused algorithm can still have an open position that gains or loses value and can still approach liquidation.
Recenter
Recentering should require new evidence: a stable range, sufficient historical movement, acceptable costs, and a new risk plan. Close or explicitly account for the old inventory before measuring the new bot. Otherwise, a shifted grid can conceal a prior loss inside a new starting point. Re-run sizing and parameter analysis instead of copying the old grid upward or downward.
Close
Close when a hard rule fires or the expected value has materially changed. Decide whether closure means cancel orders only, retain base inventory, sell to quote, or close a derivatives position. These are economically different outcomes. The confirmation screen—not the button label—should tell you what will happen.
A demo run is useful only if you rehearse the exit, reconciliation, and restart—not just the launch.
Open Bitsgap Demo Trading
A five-point exit scorecard
Use this scorecard as a review tool, not as an automated trading signal. Give one point for each statement that is true. A hard risk or security trigger can override the total and force immediate action. It turns the vague question of when to stop a grid bot into five repeatable checks.
1 point
Structure
Price has closed outside the planned range under your confirmation rule and is accepting beyond the boundary.
1 point
Equity
Total marked equity has breached the loss budget or fallen materially behind the planned benchmark.
1 point
Efficiency
Net profit per cycle no longer comfortably exceeds fees, spread, slippage, funding, and operating cost.
1 point
Market quality
Volatility shape, liquidity, or execution quality no longer resembles the setup used to design the grid.
1 point
Operations
Orders, balances, API state, or futures exposure cannot be verified reliably.
Override
Hard-stop rule
Security incident, liquidation buffer breach, delisting, inaccessible venue, or predefined risk limit takes priority.
| Score | Default interpretation | Typical next step |
|---|---|---|
| 0 | Thesis broadly intact | Continue with scheduled monitoring |
| 1 | One dimension weakening | Investigate and shorten the next review interval |
| 2 | Material mismatch developing | Pause if safe; test wait versus recenter versus close |
| 3–5 | Original setup likely invalid or operationally unsafe | Close or reduce according to the prewritten plan |
The score does not turn subjective inputs into scientific certainty. Its benefit is consistency: the same questions are asked in calm and stressful markets. Store a screenshot and short written reason at each review so you can audit whether later changes improved the process.
Three worked examples
These examples are hypothetical and omit taxes. They illustrate the reasoning process, not expected returns or recommendations.
Example A: positive grid profit, negative total result
A spot bot starts with $10,000 on an altcoin/stablecoin pair. After six weeks, completed cycles show $420 of grid profit. The bot has accumulated base inventory whose marked value is $1,080 below cost. Trading and exit costs are estimated at another $90. Approximate total P&L is therefore $420 − $1,080 − $90 = −$750, or −7.5% of starting capital.
The original plan says to close after two daily closes below support or at −7% total equity, whichever occurs first. Both have happened. The correct action under the plan is close, even though the grid-profit box is green. Waiting would be a new discretionary long position, not continued execution of the original grid thesis.
Example B: above range, but no immediate reason to market-sell
A spot BTC/stablecoin grid rises above its upper boundary and has sold nearly all BTC. Total bot equity is positive, liquidity remains deep, and the review rule requires two daily closes plus a failed retest before declaring a new trend. Only one close has occurred. The bot is mostly in quote currency, so downside exposure is limited compared with a bot stranded below range in base inventory.
The action is wait until the scheduled confirmation, not recenter immediately. If price establishes a higher range, the trader closes the old bot, records the result, and tests a new configuration. If price returns, the existing bot can resume. Waiting here is bounded by a time and structure rule.
Example C: futures grid with a shrinking liquidation buffer
A leveraged neutral futures grid has a positive realized grid figure, but accumulated fills create net long exposure during a decline. Funding has turned unfavorable and liquidation distance has narrowed below the written minimum. Margin usage also exceeds the allocation because cross margin links the position to the wider account.
This is a hard override. The bot is closed or reduced according to the derivatives plan. Adding margin merely to keep it alive is not allowed unless that contingency was sized and approved before launch. The relevant metric is survivability of the total position, not how many cycles have closed profitably.
Spot and futures bots need different exit logic
The phrase “out of range” sounds identical in both products, but the consequences differ. A spot bot generally ends with some mix of base and quote assets. A futures bot can hold a leveraged position, incur funding, consume maintenance margin, and be liquidated. That is why the answer to when to stop a grid bot must name the instrument.
| Issue | Spot grid | Futures grid |
|---|---|---|
| Below lower range | Often concentrated in base asset; unrealized loss can grow | May create net directional exposure plus margin pressure |
| Above upper range | Often concentrated in quote asset and stops cycling | Short/long/neutral mode determines exposure; do not assume flat |
| Carrying cost | Usually no funding, but capital has opportunity cost | Funding and borrow/margin mechanics can change net edge |
| Worst-case path | Asset can lose most or all of its value; custody risk remains | Liquidation can close the position before a later recovery |
| Stop action | Cancel orders and choose whether to retain or sell base | Cancel orders and verify the exchange position is fully closed |
| Key metric | Total marked equity and inventory composition | Equity, margin ratio, mark price, liquidation distance, funding |
Risk warning: Crypto assets are volatile and can lose substantial value. Leverage can accelerate losses and cause liquidation. Bot automation does not guarantee execution, profitability, or protection from exchange, API, liquidity, custody, or smart-contract failures. This page is educational, not individualized financial advice.
Safe grid bot shutdown checklist
Once the decision is made, operational mistakes can turn a sound exit into a poor one. Interfaces differ, so read the provider’s live confirmation and official documentation. Never assume that “stop,” “close,” “terminate,” and “pause” perform the same actions. Knowing when to stop a grid bot is only half the task; verifying a complete shutdown protects the result.
Snapshot the state
Record starting capital, current bot value, grid profit, unrealized P&L, fees, funding, inventory, open orders, and the reason for stopping. This preserves an audit trail before values change.
Read the confirmation options
Determine whether the platform will cancel orders only, convert base to quote, return both assets, use a market order, or close a derivatives position. Check estimated amounts and fees.
Control execution risk
For material size or thin liquidity, inspect spread and depth. A single market conversion can create slippage. Use only order methods supported by your plan and platform; do not leave an unintended exposure while optimizing pennies.
Verify at the exchange
Confirm the bot is inactive, open grid orders are gone, balances are available, and no duplicate manual orders remain. For futures, verify position size is zero when full closure was intended—not merely that the bot interface says closed.
Reconcile the final result
Export trade history if available. Calculate the final account change including fees, funding, and transfers. Note whether retained base inventory means exposure continues outside the bot.
Secure and review
If the bot used a third-party API, remove unused keys or permissions. Log the lesson, then wait before launching a replacement. A recentered grid should pass a new setup and risk review.
When a bot is not executing as expected, do not mix troubleshooting with exit judgment. First establish whether the system is actually placing and receiving orders; our frozen guide to a crypto grid bot not trading covers the operational checks. Then decide whether the strategy itself still deserves to run.
Hidden costs of stopping—and of refusing to stop
Costs of stopping
- Market-order spread and slippage
- Trading and conversion fees
- Taxable disposal or recordkeeping consequences
- Loss of future grid cycles if price re-enters
- Time out of market during redesign
- Possible withdrawal or transfer costs
Costs of continuing
- Further inventory drawdown
- Funding, borrow, or subscription expense
- Capital trapped in an idle range
- Growing liquidation or counterparty risk
- Attention spent monitoring a broken thesis
- Missed use of capital elsewhere
The existence of exit costs does not justify indefinite continuation. Compare forward-looking alternatives from the present state. The original entry price is useful for measuring performance, but it does not make future recovery more likely. Likewise, a realized loss is not economically worse than an equal unrealized loss solely because a dashboard changes its label.
Avoid closing and immediately reopening with nearly identical parameters. That often pays extra fees without changing exposure. If the range still works, waiting may be cleaner. If it does not, rebuilding around the current price requires a genuinely new thesis. To choose the grid type appropriately, compare arithmetic versus geometric grids; for downside controls, review the separate crypto grid bot stop-loss guide.
Pros, cons, and who this framework is for
Advantages of rule-based exits
- Reduces emotional decisions during fast moves
- Forces attention onto total equity, not one flattering metric
- Separates temporary inactivity from structural failure
- Makes spot and futures risks explicit
- Creates comparable records across bots
- Supports safer automation and position sizing
Limitations
- No rule eliminates gaps, slippage, or platform failure
- Confirmation can exit later than the first boundary break
- Static thresholds may need volatility adjustment
- Backtests can overfit historical behavior
- Provider metrics are not standardized
- A disciplined exit can still be followed by a price reversal
Best for
- Traders who can define a range thesis and measure total marked equity.
- Users running multiple bots who need a consistent review process.
- Spot traders who understand they may retain base inventory after stopping.
- Experienced derivatives users with explicit margin and liquidation rules.
- Anyone willing to test the entire lifecycle in demo mode before using capital.
Not for
- Anyone expecting guaranteed passive income or a hands-off money machine.
- Users who cannot tolerate the asset or platform’s loss and custody risks.
- Beginners using leverage without understanding mark price, margin, and liquidation.
- Strategies whose net edge cannot be estimated after all trading costs.
- People who intend to add funds indefinitely whenever the setup moves against them.
If you are still selecting a platform rather than managing an existing bot, compare the best crypto grid trading bots. If a grid is not suited to your market view, grid bot versus DCA bot explains the difference between harvesting a range and accumulating over time.
Frequently asked questions
When to stop a grid bot that is still showing grid profit?
Stop when total marked equity, market structure, or a hard risk rule says the original strategy is invalid—even if completed grid cycles remain positive. Grid profit can exclude an unrealized inventory loss, funding, spread, and exit costs. Reconcile the whole position before deciding.
Should I stop as soon as price leaves the grid range?
Not necessarily. One wick or brief move outside the range may reverse. Use the confirmation rule set before launch, such as closes beyond the boundary, distance measured against volatility, a failed retest, and a maximum time outside. Futures liquidation risk or a hard loss limit can require faster action.
Does stopping a spot grid bot sell all of the crypto?
It depends on the platform and the option selected. Some workflows cancel open orders and return both assets; others can sell the base asset at market. Read the confirmation screen and verify the exchange balance after stopping. Retaining base inventory means market exposure continues.
What is the difference between pause and stop?
There is no universal definition. Pause may suspend new bot actions while leaving open orders or positions intact. Stop may cancel orders, but it may or may not convert inventory or close a futures position. Check the provider’s current documentation and verify the exchange state.
Should I recenter a grid after a breakout?
Only if a new range has formed and a fresh analysis supports it. Treat recentering as a new strategy with a new starting value, backtest, cost estimate, and loss limit. Shifting a grid solely to avoid realizing an old loss is moving the goalposts.
Can a profitable grid bot become unprofitable when closed?
Yes. The final result can change because unrealized inventory is marked or sold, open orders are canceled, and spread, fees, slippage, or funding are recognized. A positive grid-profit figure alone does not guarantee positive final account P&L.
How often should I review a grid bot?
Match the review frequency to the strategy time frame and risk. A wide spot grid based on daily structure may use scheduled daily checks plus alerts. A leveraged futures grid may require much tighter monitoring of margin and liquidation distance. More checking is not automatically better; defined triggers are better.
Is a stop loss the same as stopping the bot manually?
No. A stop loss is a preconfigured trigger with platform-specific actions. Manual stopping is an operator decision and may offer different inventory choices. In both cases, confirm what happens to pending orders and remaining assets or positions.
What records should I save after closing?
Save the starting and ending equity, deposits or withdrawals, trade history, fees, funding, final inventory, reason for exit, screenshots of the closing options, and confirmation that no orders or positions remain. These records support performance review and tax reporting where applicable.
Final verdict
The best answer to when to stop a grid bot is written before launch and verified with total-equity data during operation. Close for invalidation, risk, integrity, or a completed target; pause for temporary operational uncertainty; recenter only as a newly tested strategy; and wait only with a deadline and a surviving thesis.
Never let a green grid-profit number overrule a broken range, unacceptable total loss, failed automation, or unsafe margin. The bot is an execution engine. You remain responsible for deciding whether the market still rewards the behavior it automates.
Build, monitor, pause, and close in demo mode before you decide whether real capital belongs in the strategy.
Test a Grid Bot on Bitsgap
Editorial sources and methodology
This guide was reviewed against current official or first-party documentation from BloFin, Gainium, and Pionex on September 20, 2026. Platform functions, labels, eligibility, fees, and regional availability can change; always confirm them in the live product. Hypothetical examples are original and are not performance claims.
Educational content only. Nothing on this page is financial, investment, legal, or tax advice. No result is guaranteed, and past performance does not predict future returns.