A Research Report on Dollar-Cost Averaging with Take-Profit v1.0

China A-shares · Crypto · US Equities — The Historical Validation and Final Convergence of a DCA Take-Profit Strategy

Core Conclusion (Abstract). The very same "take profit at 35%" rule is, in China A-shares, a sweet spot with an 83% win rate (simultaneously reducing drawdown and raising returns); in US equities, a net drag (a mere 20% win rate, deducting 2.6 percentage points from the mean annualized return); and in crypto, a disaster (compressing a 25.8% annualized return down to 6.7%). The optimal take-profit line for the three markets differs by an order of magnitude (35% vs. 100–150% vs. 500%+). Take-profit is not a one-size-fits-all formula, but a market-structure-specialized rule — only in China A-shares, where mean-reversion characteristics are pronounced, does it constitute a free lunch. This report validates that proposition with three independent backtests, and prescribes a deterministic execution action for each market.

I. Research Background and Core Proposition

This research originated from nearly five months of continuous operation and review of a self-built quantitative trading system. The system was built up layer by layer across multiple dimensions — daily price action, leverage structure, liquidation distribution, options sentiment, positioning structure, and macro factors — to characterize the market's real-time state. As the system grew increasingly complete, it also gradually forced out a conclusion:

Low-frequency small and mid-sized investors struggle to beat three things: macro-level global policy, black-swan information at the level of international trade, and high-frequency top-tier quantitative trading.

The first two are a crushing advantage in the temporal ordering of information — by the time policy and trade black swans have propagated down to small and mid-sized investors, prices have usually already finished reacting. The third is a crushing advantage in the processing speed of information — the structural spreads that high-frequency quant harvests at the millisecond scale are beyond the reach of manual screen-watching and experience-based decision-making. The self-built system can clearly describe "what is happening right now," but it just as honestly demonstrates: description is not prediction, and seeing clearly is not the same as possessing a sustainable excess return.

From this, the research established a pivot: since beating the market is hard, do not make beating it the goal. Do not predict direction, do not time the market, do not use leverage — and narrow the proposition from "how to beat the market" down to a question that historical data can answer:

On the premise of acknowledging that one can only capture the market β (beta), execute the single correct action within the structure of each market, via a mechanical, disciplinarily executable rule.

The concrete form of this rule is: hold cash, dollar-cost average into a broad-based index in equal amounts on each trading day, and when cumulative profit reaches a preset profit-rate threshold, sell everything, recover the capital, and start averaging in again. The work of this report is precisely to place this strategy back into real historical prices, dissect it line by line, re-validate it, and test with data its effectiveness across different market structures. It should be noted that "the difficulty of beating macro and high-frequency" is a judgment formed from the operating experience of that system; it delimits the applicable boundary of this strategy — it is designed for investors who acknowledge they possess no alpha, not for investors who claim to possess alpha.

II. Strategy Definition and Investment Targets

For a strategy to be verifiable, every rule must first be delimited to the precision of "mechanically executable day by day." The following four sections constitute the complete definition of this strategy.

2.1 Capital Structure — Three Independent Base-Currency Pools

The three pools are each settled independently in their own base currency, computing annualized returns and terminal values independently. It must be specially noted that the three do not form an exchange-rate conversion relationship — the 300,000 USDT in crypto and the 300,000 USD in US equities are not obtained by converting the 2 million RMB at an exchange rate, but rather are "equivalent amounts benchmarked against the 2-million-RMB order of magnitude." Nowhere in this document should "2 million" and "300,000" be directly converted and compared; they are three parallel experiments, not the same pool of money being ferried across three markets.

Note on definitions (derived from cross-checking the backtest logic). "Investing 300,000 into US equities and crypto over the same period" is implemented in the backtest as US equities and crypto each holding an independent 300,000 pool (each simulated separately under its own market's rules), rather than one 300,000 pool split between the two markets. This design keeps the horizontal comparison of "how the same take-profit rule performs across different market structures" cleaner — each market is backtested separately with a full-scale principal. If the actual intent were "a single USD/USDT position allocated between US equities and crypto," a separate portfolio-level joint backtest would be required (not done this round); but that difference does not affect the core conclusion of this report (the action each market should take on its own), only the way the total USD-side position is allocated.

2.2 Investment Targets — Explicit Definitions

All three are on a broad-base / blue-chip basis — not individual stocks, not small-cap altcoins, not thematic sectors. This is the foundation of the strategy: the premise of dollar-cost averaging is that the target will not go to zero over the long run and is qualified to capture β.

2.3 Execution Mechanics — Verified Line by Line Against the Backtest Logic

The strategy was handed to the backtest engine to be implemented rule by rule, and the fidelity of that implementation was verified rule by rule. The following are the precise mechanics after verification (the flow is shown in Figure 1):

1. Closed capital pool. The principal is placed in once at the start, with no external cash inflow throughout — what is invested is "cash in hand," not "future month-by-month income." This assumption means the source of DCA funds is limited and closed, and the cash recovered from take-profit will become an important source for subsequent DCA.
2. Equal-amount daily DCA. Each trading day, invest a fixed amount C = principal ÷ (DCA years × trading days per year).

- China A-shares: 2 million ÷ (5 × 244) ≈ 1,640 RMB/trading day (about 33,300/month), fully deployed over 5 years.

- US equities are converted at 252 trading days/year, crypto at 365 days/year (no days off all year) for the daily amount.

- Crypto's daily investment is split into two assets by BTC 70% / ETH 30%.

- If on some trading day the cash-pool balance is less than one share of C, invest all remaining cash.

1. Idle cash earns interest. Cash not yet invested earns interest at an annualized 2% (money-market fund / reverse-repo basis), compounded daily.
2. Take-profit full-liquidation loop (the core of the strategy). Each trading day, check: when

> current-round position market value ÷ current-round cumulative net investment − 1 ≥ take-profit threshold

holds, sell everything, return both principal and profit to the cash pool, reset current-round cumulative net investment to zero, and start DCA again from the next trading day. This mechanism can loop indefinitely: accumulate low-cost chips during the downward phase, and once a rally pushes the paper profit over the threshold, liquidate everything and bank it, then use the recovered cash to buy back day by day at C. This is the precise counterpart of the strategy's "sell early, recycle the capital."

1. No stop-loss. During paper losses, do not realize the loss but continue buying by the discipline to average down the cost and endure the drawdown (the rationale is in the data of Chapter IV).

Key verification point (one piece of precision that must be clarified). In item 4 above, the denominator of the profit rate is the "current-round cumulative net investment," which resets to zero after each take-profit full liquidation. That is, the threshold measures the paper-profit rate of "this round of purchases since the last liquidation," and not the total profit rate over the entire life cycle from initial position to now. The strategy's original phrasing, "profit on cumulative investment reaches the threshold," is precisely realized in the backtest engine as "the paper-profit rate of the current round's (reset after each liquidation) cumulative investment." This distinction determines that every loop is mutually independent and the threshold can be triggered repeatedly, and it is the precondition for the "loop" to hold. This mechanism can be replayed day by day and precisely re-checked.

On the value of the take-profit threshold. The strategy's initial setting for the take-profit threshold was 25%. The backtest (Chapter IV) shows that in China A-shares 35% is the optimal range (25% is too low and liquidates too early in a rising market, cutting into the upside); whereas in US equities and crypto, the optimal threshold must be an order of magnitude higher — which is precisely this report's most counterintuitive and most crucial finding.

An optional enhancement exclusive to China A-shares. In China A-shares, upgrading "equal-amount daily" to "valuation-weighted buying" (buy more when the valuation percentile is low, buy less or pause when it is high) can yield an additional roughly +0.3 percentage points of annualized return in the backtest. But that margin is thin and depends on the low base of the valuation percentile, and the independent audit has explicitly cautioned that "it should not be exaggerated into a robust advantage." Therefore this report's main strategy still takes "equal-amount daily DCA" as the standard (consistent with all headline data), with valuation weighting serving only as an optional refinement for China A-shares, not incorporated into the core definition. Crypto and US equities have no PE analog, so the buy side is pure DCA throughout.

2.4 Summary — Three "Do-Nots"

Do not predict direction, do not set a stop-loss, do not use leverage. This strategy compresses all active judgment into one parameter (the take-profit threshold) and one discipline (buy day by day, liquidate on meeting the threshold, buy again after liquidating). The entire effectiveness of the strategy hinges on "what value that parameter should take in each market" — and that question can be answered precisely by historical data.

III. Design Philosophy and Execution Philosophy

Before examining the data, let us make clear three design principles — they are the direct product of the judgment in Chapter I.

First, slow investing aims to reduce drawdown, not to raise returns. Faced with 2 million in hand, should one invest it all at once or gradually average in? China A-share history gives a counterintuitive answer: lump-sum did not beat DCA (the two have nearly identical average annualized returns), but lump-sum's worst window and maximum drawdown are far more brutal. The reason is that A-shares trade in ranges and mean-revert, making timing (especially "lump-sum entry at a high point") enormously risky. Therefore, spreading the 2 million into 5 years of day-by-day investment is meant not to earn more, but to ensure that even the worst starting point will not be blown through in a single stroke.

Second, take-profit aims to reduce risk, and raises returns as a side benefit only in China A-shares. The first-order effect of the take-profit loop is to "convert paper profit into banked cash, cycling the risk exposure to zero." Whether it can simultaneously raise returns depends on the market structure: in a mean-reverting market (China A-shares), selling high is likely to be followed by buying back low, so locking in profit becomes a genuine source of excess return; in a continuously rising market (US equities), selling high is followed by prices continuing to climb, so take-profit degenerates into repeatedly "selling too soon." This layer of distinction is the decisive factor of the whole piece.

Third, enduring drawdown rather than realizing a stop-loss is a bet on "mean reversion." The data (Chapter IV) show that in China A-shares, "keep holding + buy more as it falls" outperforms "cut the loss at −20%" — because a −20% stop-loss often cuts at the bottom of the pit and misses the ensuing rebound. The confidence to endure drawdown comes from the target being a broad base that will not go to zero, and from the historical characteristic of China A-shares of short bulls and long bears that eventually revert to the mean. This is a judgment, not a guarantee.

Read together, the three make the spirit of this strategy clear: it does not try to win, but tries, on the premise of acknowledging it cannot win, to turn every controllable knob to the correct position for that market's structure. Slow investing tunes drawdown, take-profit tunes risk exposure, and enduring drawdown restrains the impulse to time — three knobs, none of which depends on prediction.

IV. Historical Backtest Performance

All the statistical tables below have their figures unchanged, character for character, quoted verbatim from the three-market horizontal backtest (China A-shares 139/115 rolling windows, US equities 367/343 rolling windows, crypto a single historical path 2018→2026). Crypto's unit of account is uniformly labeled USDT; the numbers and their basis are entirely unchanged. Bold rows are the key rows of the group (sweet spot / optimum / fragile peak).

All figures in this chapter have undergone an independent blind-rewrite cross-check: a completely independently implemented backtest engine matched the main engine digit for digit across nine scenarios, with no defect in the computation path (no lookahead function, no capital leakage, and correct cost and annualized-return conventions), and a trust level judged TRUSTWORTHY.

4.1 Three-Market Headline Comparison

Three markets × no take-profit / uniform 35% / each's own optimum, 8-year holding window. Terminal-value column: for China A-shares / US equities it is the median of the rolling windows; for crypto it is the actual value of the single historical path — the two have different statistical natures and cannot be directly compared by magnitude.

Interpretation. Observe the three "uniform 35%" rows: A-shares rises from 4.0% to 5.1% (returns raised, drawdown halved at the same time); US equities falls from 9.08% to 5.76% (returns cut by about a third); crypto falls from 25.8% to 6.7% (returns cut by nearly four-fifths). One and the same rule — one market benefits, two markets are harmed. This is precisely the proposition this report sets out to prove.

4.2 Maximum Drawdown and Tail Risk — "If History Repeats, How Much Would the Worst Case Lose"

The "Drawdown-median" in the table above is the typical case; the newly computed "Drawdown-worst" in this section takes the deepest of the maximum drawdowns across each rolling window (the most brutal starting point in history). The nature of "drawdown" differs across the three: A-shares'/US equities' "worst" is a tail quantile picked out from 139/367 historical starting points (statistically meaningful); crypto's "drawdown" is the value that actually occurred on the single historical path (a different starting point could yield an entirely different result — this is "single-sample"-level evidence). The numbers can be placed side by side, but their confidence is not of the same order of magnitude.

China A-shares · 8-year/10-year windows · median drawdown vs. worst drawdown

US equities · 8-year/10-year windows · median drawdown vs. worst drawdown

One new finding: US-equity take-profit provides almost no protection against the "worst case." A-share take-profit at 35% compresses the worst drawdown from -49.0% to -38.2% (a compression of 10.8 percentage points); US-equity take-profit at 35% only compresses the worst drawdown from -55.3% to -54.4% (a compression of 0.9 percentage points), and take-profit at 100%/150% is even completely tied with no take-profit (both -55.3%, both crashing into the window starting 2001-03). The reason is that the most brutal windows in US-equity history are the rapid crashes of the 2000–2002 dot-com bubble and the 2008-financial-crisis type — DCA and the take-profit loop are blown through before they can accumulate a sufficient profit buffer; in a "comes fast, falls deep" scenario, the take-profit rule is far less effective than in A-share-style range oscillation. "Take-profit can cushion the worst case" is itself also an A-share-specialized experience.

Crypto · full candidate take-profit-line scan (drawdown is the actual value of the single path · unit USDT)

It must be emphasized again: the "max drawdown" in the table above is the value that actually occurred on this one historical path (2018-01-01→2026-08-08), not the worst quantile of a probability distribution. The uniform-rule 35% tier's drawdown of only -11.0% looks "safe," but that is really because it triggered take-profit 12 times on this path, locking in only small profits almost the entire way — a different starting point may not yield the same.

4.3 Full Candidate Scan of the Take-Profit Line — Locating the Optimal Threshold

China A-shares · 10-year window candidate scan

The sweet spot for China A-shares clearly falls at 35%: an 87% win rate, 1.70 percentage points of additional annualized return, drawdown compressed from -46% to -17%, and losing windows reduced to zero. 25% liquidates too early; 50% leaves too much drawdown.

US equities · full candidate scan (8-year/10-year windows)

US equities' "optimum" is a trap. Selecting by median annualized return, the 8-year-window optimum is take-profit 100% (9.78% > the 9.08% benchmark), and the 10-year-window optimum is take-profit 150%. Yet once you switch to the mean annualized return (expected-value basis): the mean for take-profit 100% is 7.44%, which is actually lower than no take-profit's 7.97%, and the window-by-window paired win rate is only 17%. This shows that in US equities, take-profit merely compresses the variance of the distribution (shaving the peaks while filling the troughs), flattening the long right tail of the bull market without raising the expected return. The higher median is an illusion of "the distribution being squeezed flat," not "greater profitability." This is not a free lunch, but trading expected value for stability of the typical result.

Crypto · extended scan (range 50%→2000%, data in the 4.2 scan table). Crypto does not converge monotonically, but the scan reveals a genuinely existing yet extremely fragile internal peak: the 1000% tier has CAGR 30.5% and drawdown -50.3%, simultaneously superior to buy-and-hold (25.8%/-74.7%), but on both sides of the peak it drops off a cliff — one tier lower (750%) is only 25.5%/yr, and one tier higher (1500%) plunges straight back to 25.8%/yr (equivalent to no take-profit at all, triggering 0 times).

This peak is luck rather than a parameter, on two grounds. First, the 1000% threshold triggered only once on this path, on 2021-11-08 — precisely near the closing top of Bitcoin/Ethereum in the 2021 bull cycle, with the take-profit selling exactly at the cycle ceiling and dodging the ensuing -75% bear market of 2022; this is "a single historical path where one exit timing happened to hit the luck right," not a smooth return-drawdown trade-off curve. Second, when the independent audit supplemented the test with a finer grid at the "1250%" tier, the CAGR was even higher (36.6% > 30.5%) — because 1250% was never triggered throughout the entire 2021 cycle (the gain fell short of 12.5×), dragging all the way to 2025 before exiting, by which point the exit timing had shifted to a different cycle. Even "which tier is optimal" drifts with the coarseness or fineness of the scan grid; there is no stable, reproducible "optimal take-profit line." Chasing that number itself constitutes overfitting.

4.4 Core Conclusions

1. The take-profit line cannot be one-size-fits-all. The same 35% line is a sweet spot in China A-shares (83% win rate), a net drag in US equities (only 20% win rate, deducting 2.6 pp/yr), and a disaster in crypto (25.8% compressed to 6.7%). The three optimal take-profit lines differ by an order of magnitude (35% vs. 100–150% vs. 500%+).
2. The China A-share conclusion that "take-profit = reduces risk without adding return" cannot be extrapolated wholesale. China A-shares: take-profit simultaneously reduces drawdown (-42%→-17%) and raises returns (4.0%→5.1%), a win-win, because mean reversion makes "locking in profit" a genuine excess return. US equities: the so-called "optimal" take-profit holds only on the median basis; on the mean basis it is actually worse (7.44%<7.97%), with a paired win rate of only 17% — take-profit merely compresses variance without raising the expectation. Crypto: take-profit is pure return-for-drawdown, not disguised as adding return.
3. The crypto take-profit-line scan exposes the fragility trap of a single historical path. The trigger timing of the 1000% tier happened to hit near this cycle's high, making CAGR overtake buy-and-hold, but the neighboring tiers on both sides drop off a cliff, and the "optimal point" drifts with the grid — judged as luck rather than a trustworthy parameter.
4. Statistical robustness differs by three orders of magnitude. China A-shares' 139 windows and US equities' 367 windows (both trustworthy statistical distributions) versus crypto's mere 1 path — crypto's conclusion is by nature "single-sample" and cannot be treated as equivalent to the distribution-based conclusions of the other two markets.
5. Take-profit's protective power against the "worst case" also differs vastly across the three markets. A-share take-profit at 35% compresses the worst drawdown by 10.8 percentage points; US-equity take-profit at 35%/100%/150% provides almost no protection against the worst drawdown (compressing by 0.9 percentage points or even breaking even) — in rapid-crash markets, the take-profit loop is blown through before it can accumulate a buffer.

V. Final Convergence — Three Markets, Three Deterministic Actions

This chapter no longer lists probability ranges, no longer distinguishes "8-year vs. 10-year," and no longer enumerates multiple sets of annualized numbers. The previous four chapters have completed all the argumentation; here we answer only one question: exactly which action each market should execute by.

China A-shares (CSI 500 Total Return)

Deterministic action: equal-amount daily DCA + take-profit 35% full-liquidation loop + no stop-loss.

This is the only win-win among the three markets: take-profit simultaneously halves the drawdown and lifts the returns, because A-shares' range oscillation and mean reversion make "buying low again after selling high" a genuine source of excess return. 35% is the sweet spot and should not be adjusted — lower will cut into the upside too early, higher will leave too much drawdown. During the downward phase, keep holding and buy more as it falls. One honest caveat must be stated up front: this excess return is equivalent to betting that "China A-shares continue their short-bull-long-bear, mean-reverting pattern"; if China A-shares in the future walk a US-style sustained long bull, take-profit will repeatedly sell too soon and substantially underperform. This is a judgment, not a guarantee.

Crypto (BTC 70% / ETH 30%)

Deterministic action: only DCA + long-term holding; do not borrow the A-share-style take-profit.

In crypto, that A-share 35% take-profit compresses the 25.8% annualized down to 6.7%, evaporating nearly four-fifths of the return. The only high take-profit line (above 500%) that could "balance return and drawdown" is itself not robust, drifts with the scan granularity, and is luck rather than a parameter — it cannot be trusted. Reducing crypto risk should rely on a position cap (i.e., the amount discipline of 300,000 USDT) and annual rebalancing, not on take-profit. A caution: all conclusions for crypto are built on a single historical path, the weakest strength of evidence, closer to a cautionary sample than a trustworthy probability distribution.

US equities (S&P 500 Total Return)

Deterministic action: pure DCA + buy-and-hold, no take-profit.

In US equities, take-profit is a negative-sum operation — the so-called "optimum" holds only on the median basis, is actually worse on the expected-value basis, has a paired win rate of only 17%, flattens the long right tail of the bull market, and provides almost no protection against rapid crashes. The correct posture for US equities is precisely to buy and hold a broad base for the long term, abandoning take-profit entirely.

Long-Term Behavior (Cross-Market Guiding Principle)

Deterministic judgment: the only reliable edge for low-frequency small and mid-sized capital is not "outperforming," but "not being eliminated + capturing, with discipline, the β that matches one's own market structure."

Active rules like take-profit are market-structure-specialized — they constitute a free lunch only in mean-reverting China A-shares; in a trending market (US equities) and a high-volatility one-directional growth market (crypto), the same rule turns into self-harm. Hence there exists no "universal formula" that can be applied across markets. Acknowledging that one cannot beat macro black swans and high-frequency quant means: do not predict, do not time, do not use leverage, do not chase a single fragile optimal parameter, and place all one's energy on "selecting, for each market, the action that matches its structure."

This is the final judgment converged upon after five months of continuous operation of the self-built quantitative trading system: what is reliable is not a more ingenious prediction, but a more honest discipline.

VI. Honest Boundaries

The following are preserved item by item, not pruned for the report's appearance.

• Past ≠ future. All conclusions derive from the historical intervals of 2007–2026 (one and a half bull-bear cycles in China A-shares), 1988–2026 (US equities), and 2018–2026 (crypto), and do not imply the future will repeat.
• The rolling windows are highly overlapping. China A-shares' 139/115 and US equities' 367/343 windows roll month by month, with adjacent windows sharing the vast majority of trading days; the truly independent sample size is far smaller than the surface number, and the significance is weaker than it appears.
• Crypto is not statistically robust. Mainstream-exchange Bitcoin/Ethereum data only start in 2017, and 2018→2026 is the only complete window that can be covered, with no month-by-month rolling done — this is "single-sample"-level evidence.
• Cost assumptions are not uniform across the three markets. China A-shares use 2bp buy / 5bp sell (locked by the audit); US equities carry over the same A-share figures, uncalibrated to the actual US commission-and-tax regime; crypto's 5bp buy / 10bp sell is a working assumption drafted this round, not yet finalized, and the real exchange fee rates may differ substantially.
• Only China A-shares' buy side does valuation weighting (and with a thin margin); US equities/crypto are both pure DCA, with no verification of whether valuation timing is effective in those two markets.
• None of the three markets sets a stop-loss (carrying over enduring drawdown); this round did not reopen that decision.
• None of the three markets did an exchange-rate-converted version; each is priced independently in its own base currency, aligned only in principal magnitude, and does not constitute a cross-market comparison of the same pool of money.
• Crypto's 1000% tier is luck rather than mechanism, and the "optimal point" drifts with the scan grid — this is an in-sample parameter-tuning bias of a single path; do not misread it as "a precisely calibrated number."
• The US-equity "optimal" take-profit line has a median/mean-basis divergence — whether to adopt it depends on whether one weights the stability of the typical result or the maximization of the long-term expectation; the two cannot be had at once.
• This report is a research framework and historical statistics, and does not constitute investment advice. Decisions and their consequences are your own.