The Case of Bitcoin, and the Epistemology of Its Parameters
Working paper · Methodological research June 2026
Abstract
This paper examines a class of capital-deployment methods for building positions in assets that exhibit endogenous cyclicality, using Bitcoin as its case. Bitcoin’s protocol embeds a deterministic supply-decay mechanism — the block subsidy halves every 210,000 blocks, approximately four years — producing a price history with a non-sentiment-driven periodic structure. This institutional fact is the foundation on which the method rests.
We begin from the two classical frameworks, dollar-cost averaging (DCA) and value averaging (VA) [1][2], and identify a shortcoming in each when applied to an asset with a directional downside prior and extreme volatility: DCA discards price information entirely, while full VA demands unbounded capital in a sustained decline. As a synthesis we construct a continuous weighting function in which per-period deployment scales smoothly and monotonically with the price’s deviation from a reference anchor, with an aggressiveness parameter k governing contrarian intensity and a cap providing tail circuit-breaking.
The paper’s second and principal contribution is epistemological. We argue that the parameters of such a framework are not technical settings but frozen predictions — priced encodings of beliefs the investor cannot verify. We decompose any such belief into three layers (world model, object, self) and demonstrate a falsifiability asymmetry: these layers report back at radically different speeds, and the most aggressive parameter necessarily rests on the least testable layer. We further identify an irreducible self-referential defect: a rule designed to eliminate judgment is itself an act of judgment. We conclude that the reserve — capital withheld from the model entirely — is the only operationally honest response to structural ignorance.
All exposition is symbolic and dimensionless. The paper presents a methodology, not a position.
Keywords: Bitcoin; halving cycle; dollar-cost averaging; value averaging; contrarian allocation; falsifiability; execution risk; epistemology of models
1. Introduction
Staged accumulation is among the most common strategies available to the individual investor facing uncertainty. Its simplest form is dollar-cost averaging: invest a fixed sum at fixed intervals, without timing. Its popularity derives largely from its avoidance of a hard problem — the investor need not judge when the market has bottomed; he need only execute.
Since Constantinides (1979) [1], however, the literature has held DCA to be suboptimal in expected-utility terms relative to lump-sum investment. The core of the critique is that DCA is a deterministic strategy that does not update on new information, and therefore cannot concentrate deployment when the asset is cheap or withdraw when it is dear.
But that critique presupposes i.i.d. returns and the absence of exploitable structural information. When the underlying asset possesses an endogenous, non-sentiment-driven periodicity, the presupposition fails — the price’s position relative to an anchor itself carries information, and “buying more when low” ceases to be a behavioral bias and becomes a rational response to structure.
Bitcoin supplies precisely such an asset. Its supply side decays deterministically and predictably by protocol (Section 2), not by central-bank discretion or market sentiment. This institutional feature makes “price relative to cycle position” a weak but genuine structural signal.
The question this paper takes up is therefore not the settled one — is DCA optimal? — but rather: when an asset carries weakly exploitable cyclical information, how can one recover partial use of price information while retaining DCA’s executional discipline?
And then, more uncomfortably: what exactly is one believing when one chooses the parameters that govern that recovery?
Structure. Section 2 presents Bitcoin’s halving mechanism and bounds it as a weakly exploitable prior. Section 3 reviews the three relevant frameworks. Section 4 constructs the weighting function axiomatically. Section 5 — the epistemological core — analyzes what the parameters actually encode. Section 6 treats execution. Section 7 integrates. Section 8 states the limits.
A note on method. All quantities are expressed as proportions of total investable capital C, and all prices in normalized form p = P/Pref. This serves two ends: it renders conclusions independent of position size, presenting a transferable methodology rather than a case; and it forestalls the misreading of a framework-selection discussion as investment advice on a particular asset. This paper is not investment advice.
2. Institutional Background: The Halving Mechanism
This section introduces only those institutional facts that the argument requires. It is deliberately not an encyclopedic treatment; each fact is stated because a later argument depends on it.
2.1 Issuance and the halving rule
Bitcoin launched in January 2009. New coins enter circulation through the block subsidy: each miner producing a block receives a fixed quantity of new bitcoin. The protocol specifies that the subsidy halves every 210,000 blocks. Because the difficulty adjustment (every 2,016 blocks) anchors mean block time near ten minutes, 210,000 blocks corresponds to approximately four years — this is the protocol-level origin of the “four-year cycle.”
Four halvings have occurred [3]:
| Halving | Date | Block height | Subsidy (BTC) |
|---|---|---|---|
| First | 2012-11-28 | 210,000 | 50 → 25 |
| Second | 2016-07-09 | 420,000 | 25 → 12.5 |
| Third | 2020-05-11 | 630,000 | 12.5 → 6.25 |
| Fourth | 2024-04-19 | 840,000 | 6.25 → 3.125 |
| Fifth (projected) | ~2028 | 1,050,000 | 3.125 → 1.5625 |
Table 1. Dates drift with realized block rates; future halvings can be estimated but not fixed.
Successive halvings continue until the full 21 million supply is mined (circa 2140). This fixed cap combined with a decaying issuance schedule is the code-level root of Bitcoin’s scarcity narrative — it distinguishes the asset from fiat currencies subject to discretionary issuance. As of the fourth halving, over 19.68 million coins had been mined (above 93% of total supply), and the annualized new-supply rate had fallen below 1% [3].
2.2 Why the supply shock appears in log space
The transmission of a halving to price has a property that is routinely misunderstood: it acts on the percentage rate of new supply, not on an absolute quantity. Each halving cuts the growth rate of supply in half — a multiplicative, not additive, shock.
The methodological consequence is direct: cyclical structure is legible in log-price space and drowned in linear price space, where the sheer change in price magnitude across cycles swamps the pattern. This is why the weighting function developed in Section 4 takes relative price p = P/Pref — a ratio, i.e. a translation in log space — as its argument rather than an absolute price difference. Deployment intensity should respond to how many percent cheaper, not how many dollars cheaper.
2.3 The cycle, and the “bottom before the halving” heuristic
Following each halving, price has reached its cycle peak with a lag of typically six to eighteen months, not at the halving instant [3]. At the symmetric end is the cycle trough: a heuristic widely cited in trading communities holds that price tends to bottom roughly 500 days before a halving.
If the heuristic holds, then time-distance to the next halving is itself a crude indicator of cycle position — an external reference for setting the anchor Pref and for judging whether a low is near. But it must be stressed that this is an empirical observation, not a protocol guarantee, and its timing has varied substantially across cycles.
2.4 The cycle as a weakly exploitable prior: three cautions
This paper positions the halving cycle as a weakly exploitable prior, not a deterministic forecast. The positioning is decisive: it is precisely what dictates the choice of moderate use of price information over aggressive betting on timing.
The sample is minuscule. Only three complete cycles are observable (the fourth is in progress). Any “regularity” rests on vanishingly few degrees of freedom. Meynkhard (2019) finds halvings associated with systematic increases in market value [4]; empirical examination of the 2012, 2016, and 2020 halvings nonetheless finds the timing of peaks and troughs inconsistent across cycles [5].
Descriptive models fail out of sample. Models mapping scarcity directly to price (Stock-to-Flow and kin) do not outperform a naïve current-price forecast under formal out-of-sample testing, and should not be mistaken for prediction engines [6].
The structure is mutating. Spot ETF approval in January 2024 and the entry of institutional capital may dampen volatility and lengthen or distort the existing cycle form; post-fourth-halving price behavior already differs markedly from the prior three [7]. Even if the prior held historically, its strength under the current structure may be decaying.
Section summary. The halving mechanism furnishes a prior that is real but fragile: real, because it originates in protocol code rather than sentiment; fragile, because the sample is tiny, the descriptive models fail out of sample, and the market structure is being rewritten by institutionalization. A robust framework should exploit this prior without depending on it being true — which is the root justification for the “controlled contrarian” design that follows.
3. Theoretical Background: Three Frameworks
3.1 Dollar-cost averaging and its critics
DCA’s mathematics are transparent. Investing a fixed sum each period buys more units when prices are low and fewer when high; the resulting average cost is the harmonic mean of the period prices, which never exceeds the arithmetic mean. This “automatic buy-low” property is routinely cited as DCA’s virtue.
Constantinides (1979) nonetheless proves its suboptimality in a mean-variance frame [1]. Brennan, Li, and Torous [8] extend the critique: DCA makes the investor’s holdings at any time depend on his initial portfolio, contradicting the rational-expectations principle that an optimal portfolio should depend only on current information. Milevsky and Posner (1999) re-examine DCA with continuous-time option-pricing tools, characterizing its terminal payoff as an Asian-option structure [9].
Against this efficiency critique stands Statman’s (1995) behavioral defense [10]: DCA’s pre-committed mechanical rule mitigates the emotional pressure and regret-aversion an investor suffers amid volatility. The substance of this defense is that DCA’s value lies not in expected return but in executability — a theoretically superior strategy the investor cannot adhere to may realize lower returns than an inferior one he can.
The design that follows lives inside this tension throughout: moving toward the use of price information without surrendering disciplined executability.
3.2 Value averaging: the opposite pole of information use
Edleson’s value averaging (first published 1988; book 1993) [2] occupies the pole of maximal price-information use. It fixes not the invested amount but the target growth path of portfolio value, then back-solves each period’s required contribution: a price decline drops portfolio value below the path, requiring a larger contribution; a rise requires less, or a sale. VA thus implements “buy more when it falls” natively, and continuously. In most historical backtests its cost control dominates DCA’s.
But VA carries a limitation fatal to the present case: in a sustained decline, the contribution required to maintain the target path grows monotonically and can rapidly breach any fixed budget. For an investor operating under a hard capital cap with no external replenishment, full VA is inapplicable — it tacitly assumes funds can be topped up without bound. This observation is the origin of the controlled weighting function constructed below: retain VA’s contrarian spirit, impose budget feasibility.
3.3 The Wyckoff schematic: bounding a cognitive tool
Unlike the preceding two, the Wyckoff accumulation model (1930s) is not a capital rule but a description of market structure [11]. It portrays large capital accumulating quietly within a ranging base, shaking out retail holders through repeated oscillation; its stage markers — selling climax, secondary test, the “spring” (a false breakdown), sign of strength — are widely invoked in trading practice. Its observation that the selling climax often marks the range low runs in the same direction as this paper’s contrarian logic and is genuinely suggestive.
But deploying it as a real-time timing signal contains a fundamental methodological defect. The pivotal “spring” can only be confirmed after price successfully reclaims the range and breaks upward — in the moment itself, a spring is formally indistinguishable from a genuine breakdown. It is a label that is identifiable ex post and undecidable ex ante.
The paper therefore restricts the Wyckoff frame strictly to a tool for understanding market structure, and excludes it from the trigger conditions of any deployment rule. This restriction is itself a statement of the paper’s methodological stance:
A robust deployment framework must not derive its validity from a real-time determination of the current structural position.
4. The Continuous Weighting Function
4.1 Setup and notation
Let total investable capital be C. Let Pref denote an exogenously chosen reference anchor (a “neutral” price), P the current price, and define the normalized price p = P/Pref. Let b be the baseline deployment intensity (the per-period sum at p = 1, itself expressible as a fraction of C).
We seek a per-period deployment function I(p) satisfying the following axioms:
- (A1) Monotonicity. I is non-increasing in p — the cheaper the asset relative to the anchor, the larger the deployment.
- (A2) Normalization. I(1) = b — at the anchor, the function reduces to baseline DCA.
- (A3) Continuity. I is continuous, avoiding the discontinuous jumps and “gapped intervals” of discrete tiers.
- (A4) Tunable aggressiveness. There exists a parameter k controlling the curvature of contrarian amplification, with k = 0 collapsing to price-independent DCA.
- (A5) Budget feasibility. An upper bound exists such that per-period deployment is bounded, so that a sustained decline cannot violate the capital constraint — the point of departure from full VA.
4.2 The power law
The simplest form satisfying (A1)–(A4) is a power law:
$$I(p) = b \cdot p^{-k} = b \cdot \left(\frac{P_{\text{ref}}}{P}\right)^{k} \tag{1}$$where k ≥ 0 is the aggressiveness parameter. Its methodological content is legible in its limiting behavior:
- k = 0: I ≡ b. Price information is discarded entirely; the framework degenerates to classical DCA — precisely the “deterministic, non-updating” strategy Constantinides attacked.
- k = 1: deployment is inversely proportional to price, so that amount × price = constant units — equivalent to buying a fixed quantity each period.
- k ≥ 2: the function is convex; deployment accelerates in deep declines, approaching VA’s contrarian intensity while remaining bounded by (A5).
Thus k is a continuous dial: one end (k = 0) is DCA, the other (large k) approaches a controlled VA.
The essence of this framework is to open a continuously tunable middle territory between DCA and VA.
4.3 Dimensionless deployment intensity
Table 2 gives normalized intensity I/b = p−k as a function of normalized price. It is dimensionless and independent of any monetary amount.
| Normalized price | k = 0 | k = 1 | k = 2 | k = 3 |
|---|---|---|---|---|
| p = 0.60 | 1.00 | 1.67 | 2.78 | 4.63 |
| p = 0.70 | 1.00 | 1.43 | 2.04 | 2.92 |
| p = 0.80 | 1.00 | 1.25 | 1.56 | 1.95 |
| p = 0.90 | 1.00 | 1.11 | 1.24 | 1.37 |
| p = 1.00 | 1.00 | 1.00 | 1.00 | 1.00 |
| p = 1.10 | 1.00 | 0.91 | 0.83 | 0.75 |
| p = 1.20 | 1.00 | 0.83 | 0.69 | 0.58 |
| p = 1.40 | 1.00 | 0.71 | 0.51 | 0.36 |
| p = 1.60 | 1.00 | 0.63 | 0.39 | 0.24 |
Table 2. Normalized deployment intensity. p < 1 is the relatively cheap region; p > 1 the relatively expensive one. Higher k lifts the curve more steeply below the anchor and suppresses it more deeply above.
4.4 Tail circuit-breaking: restoring budget feasibility
The power law diverges as p → 0 for k ≥ 1, violating (A5). To restore feasibility, impose a per-period cap Imax = m · b (with m > 1 the cap multiple):
$$I(p) = \min\left\{ b \cdot p^{-k},\; m \cdot b \right\}, \qquad I(p) = 0 \;\text{ for } p > \theta \tag{2}$$where θ > 1 is the high-side valve: deployment is suspended when price stands materially above the anchor.
The two truncations in (2) serve distinct ends. The cap m·b governs the downside tail (preventing unbounded deployment in a crash); the valve θ governs the upside (preventing sustained buying at elevated prices).
A subtlety of first-order importance: the cap’s engagement price is jointly determined by m and k. Solving p−k = m gives engagement at
$$p^{*} = m^{-1/k} \tag{3}$$This means the cap is not active across the whole decline — it intervenes only below p, while deployment intensity everywhere above p remains governed by k alone.**
The design failure this invites is subtle and easy to commit: if m is set too tight while k is large, the cap engages in the moderate decline region and suppresses deployment precisely where one most wanted to add — producing the counterintuitive behavior “bold in moderate declines, timid in extreme ones.” Conversely, widening m so that engagement retreats to the extreme tail allows k’s contrarian amplification to operate fully across the normal decline, with the cap relegated to its proper role as pure tail insurance.
m and k cannot be set independently. They must be jointly calibrated so that deployment intensity is monotone in price and caps only in the intended extreme.
5. What the Parameters Actually Are
This section is the paper’s principal contribution, and it is uncomfortable.
Equation (2) contains four parameters: baseline b, anchor Pref, aggressiveness k, cap m (with valve θ). They are not technical dials. Each encodes an unverifiable belief about the world, and prices it.
5.1 Three layers of any judgment
When an investor says I believe the price will fall to X, that sentence smuggles three categorically different things inside it. Conflating them is the origin of most error.
- Layer 1 — World model. How does the world work? Do supply shocks propagate to price? Do systems with feedback and delay oscillate? Is periodicity a property of the world at all?
- Layer 2 — Object. What is this particular thing? Is Bitcoin still the asset the model assumes? Has the ETF era changed what it is?
- Layer 3 — Self. Who am I? What drawdown can I actually bear, as against what I claim I can bear? What premium will I pay for the possibility that I am entirely wrong?
Any judgment is the product of all three.
The value of separating them emerges only after failure:
When the strategy loses, the question is no longer “was I wrong?” — that question has no answer and no use. The question is: which layer was wrong?
If Layer 3 failed — the investor could not hold the position he swore he could — the model was sound and the human was not. The remedy is smaller size, not a rewritten model.
If Layer 2 failed — the cycle is real but the asset has changed — the method survives. Change the parameters or the object.
If Layer 1 failed — the cycle never existed and three samples were coincidence — everything downstream must be demolished.
Three failures, three remedies. Absent the decomposition, all one obtains is a useless sentence: I got it wrong.
5.2 The falsifiability asymmetry
This is the observation that ought to govern parameter selection, and it is routinely absent from the literature.
The three layers report back at radically different speeds.
| Layer | When you learn you were wrong |
|---|---|
| Self | Months. When the feared price actually prints, you discover what you are. The body does not lie. |
| Object | One to two years. Fund flows, holder structure, realized volatility are all observable. |
| World model | Possibly never. |
Why never? Because the world model requires samples, and Bitcoin’s cycle produces one sample every four years. To confirm “halvings generate cycles” at any respectable confidence would require ten cycles — forty years.
And capital must be allocated now.
From this follows the paper’s most uncomfortable conclusion:
The most aggressive parameter rests on the least verifiable layer.
k encodes deep declines are worth loading into. The warrant for deep declines will come is the cycle. The warrant for the cycle is three samples.
Three.
Any honest statistician, presented with n = 3, will say: you have demonstrated nothing. He is correct. And the investor commits regardless.
Therefore, when this framework fails, the likeliest cause is not that the investor misjudged himself (he would learn that quickly), nor that he misread the object (that is correctable within a year).
The likeliest cause is that the world model was wrong — and he will never accumulate the samples to confirm it.
He will lose, and he will not know why.
That asymmetry — not volatility, not drawdown — is the real risk the framework carries.
5.3 Parameters are frozen predictions
There is a position-dependent blindness at work here, and it is produced by competence rather than its absence.
An experienced investor is excellent at reading worldviews out of other people’s models. Handed a business plan assuming a 45% gross margin holds for three years, he is not reading the number — he is reading how that founder understands his own moat. He reads a discount rate and infers a private theory of risk.
But when he sets his own k, he experiences it as a technical choice.
Not from stupidity. Because in that seat he is the designer, not the auditor.
The auditor reads parameters. The designer writes them. And in the act of writing, one cannot see what one is writing.
Turn the audit inward and the parameters resolve into a portrait:
- Pref — the price called “fair.” It was not computed. It was believed.
- k — a belief that extremes pay; a whole theory of how the world rewards courage.
- m (the cap) — but not full belief. A brake installed against being wrong about the extremes.
- The reserve (Section 7) — and not full belief in any of it. Capital withheld against the falsity of the entire theory.
Set side by side, these four say: This investor believes in the cycle, but not completely. He will add into fear, but not to the point of losing his footing. He has priced his conviction — and he has also priced his error.
This is a more accurate description of him than anything he could write about himself, because it was written with capital.
One can lie in a self-description. One cannot lie in one’s parameters.
And the corollary, which is the operational payload of this section:
The claim “my model does not require me to predict the bottom” is false. The prediction has not been eliminated. It has been relocated into Pref and k.
Parameters are frozen predictions.
5.4 The self-referential defect
There is a hole here that cannot be patched, and intellectual honesty requires stating it rather than concealing it beneath the mathematics.
The stated purpose of the framework is to spare the investor from judging in moments of panic — because he does not trust the version of himself that panics.
But who wrote the rules?
He did. With the same brain.
He merely used it at a calmer hour, froze the output, and set it to govern the man who arrives later.
The structure is self-referential: an element inside the system (the investor) generates a rule constraining that system (including himself).
And the Gödelian fissure opens exactly here:
- “I should not judge while panicking” — is itself a judgment.
- “My rule does not require me to predict the bottom” — but the beliefs that a bottom is worth predicting and that it lies in a particular region are already encoded in Pref and k.
There is no such thing as a judgment-free rule, because designing is judging.
The loop is not escaped. It is merely drawn larger.
And drawing it larger may be all that is available.
6. Structural Problems in Execution
A framework coherent in theory can still fail in execution. Two structural problems are commonly overlooked and materially alter behavior.
6.1 Decision frequency and sampling bias
If deployment decisions are taken only at sparse fixed times (say, monthly), while the asset trades continuously with high volatility, the result is systematic sampling bias: the extreme lows within a period will very often fall between decision points and be missed entirely.
The consequence is that realized entry prices skew toward the interval mean or above — exactly contrary to the design intent of “deploy heavily at extreme lows.” Low-frequency discrete decisions blunt the very downside tail the weighting function exists to exploit.
6.2 Two layers, two frequencies
The remedy is to decompose deployment by time sensitivity and assign each layer its own frequency:
| Layer | Trigger | Frequency | Time sensitivity |
|---|---|---|---|
| Baseline deployment | Manual, at period open, priced off the curve | Low (periodic) | Low — averaging is insensitive to exact timing |
| Deep-dip capture | Pre-placed limit orders | Real-time, automatic | High — must capture transient lows |
Principle: let the time-insensitive component enjoy the executional simplicity of low frequency, and grant the time-sensitive component real-time capture.
6.3 The tension between discrete triggers and a continuous function
The idealized function (2) presumes execution at arbitrary prices with the corresponding intensity; real conditional orders rest only at discrete levels. A pragmatic reconciliation is to have the low-frequency layer approximate continuity (evaluating the function at the realized price each period) while discrete conditional orders cover the tail (pre-placed additions at key levels below).
The superposition is a piecewise approximation to the theoretical curve. It sacrifices some mathematical purity for executability. Acknowledging and explicitly managing this approximation is more honest than pretending a continuous function can be executed perfectly.
7. The Integrated Framework, and the Reserve
7.1 Components
| Component | Form | Methodological function |
|---|---|---|
| Capital split | Main α·C / reserve (1−α)·C | Main capital deploys under the model; the reserve is hard-locked against the model being false |
| Baseline layer | I(p) = min{b·p−k, m·b}, p ≤ θ | Low-frequency, continuous weighting, approximate use of price information |
| Deep-dip layer | Pre-placed conditional orders at discrete levels | Real-time standby; captures lows the low-frequency layer misses |
| High valve | Suspend and accrue when p > θ | Suppresses chasing above the anchor; deferred funds release on decline |
| Tail circuit-breaker | Cap m·b + locked reserve | Guarantees bounded deployment and non-exhaustion of capital in a sustained decline |
7.2 Qualitative behavior under three regimes
- Deep decline, then recovery. The weighting function and dip orders activate fully; deployment concentrates at the lows; average cost falls markedly. The framework’s best case.
- Sustained elevated grind, no deep decline. The valve triggers repeatedly, intensity is suppressed, and much of the main capital remains unspent. The framework restrains itself — at the cost of significant under-allocation for the cycle.
- Extreme crash. Cap and reserve engage in sequence, establishing position at extreme lows under a bounded-deployment constraint. The framework’s tail-protection case.
The trade-off is explicit: the framework accepts under-allocation in the grind regime to purchase low-cost accumulation in the decline regime and bounded risk in the crash regime. Whether that trade is worthwhile depends entirely on the strength of the investor’s prior about downside — which is to say, it returns to k. The framework does not dissolve that judgment. It renders it explicit, tunable, and auditable.
7.3 The reserve as materialized ignorance
The reserve deserves separate treatment, because it is not merely a risk-management device. It is the operational form of an epistemological commitment.
Section 2.4 conceded a dark region — the possibility that the cycle prior is simply false, and that no attainable sample could reveal this in time. What should a sincere agnostic do about that?
The common answer is to write an essay arguing that the world is unknowable.
The answer proposed here is: withhold capital.
The reserve is not provisioned for a modeled bad case. That case is already inside the model; it is not the unknown. The reserve is provisioned for:
- the case in which the cycle framework is false at the root;
- the case in which the asset is not the kind of thing the model assumes;
- the cases that cannot yet be named.
The reserve is the dark region, converted into something with a price.
This is the framework’s most honest component, because it is not a claim but a cost — real, paid, carrying an opportunity cost. It is the one place where the investor’s stated uncertainty is made expensive to him.
And it yields a diagnostic that generalizes beyond finance:
When a person’s arguments and a person’s allocations disagree, trust the allocations.
Arguments are free. Allocations are not. A belief one has paid for is the only belief one actually holds.
An investor who argues that the cycle is certain, yet retains a reserve, has already conceded — in the only currency that counts — that he might be entirely wrong.
8. Limits and Boundaries
The prior is fragile. The framework’s coherence depends on a weakly exploitable endogenous cycle. As Section 2.4 established, the statistical evidence rests on a minute sample and may be attenuated by structural evolution. Should the prior fail, the framework degenerates to a price-tilted DCA, and its advantage over DCA evaporates with it.
The anchor is exogenous. Pref is supplied by the investor; the framework offers no method for estimating it. Anchor error, amplified by k (the elasticity ∂ln I / ∂ln Pref = k), can distort deployment systematically. This is the framework’s most fragile input, and it is fragile in exact proportion to how aggressive k is.
Non-optimality. No claim is made of optimality in any utility sense. Constantinides’ critique applies here in part: this remains a rule-based strategy, merely one that exploits one additional dimension of information. Its defense, with Statman, rests on executability rather than expected return.
Counterparty and liquidity risk. Pre-placed conditional orders require capital resident at an execution venue, introducing counterparty exposure; transient illiquidity may cause realized fills in the dip layer to deviate from intent.
The self-referential defect is not resolved. Section 5.4 stands unanswered. It is stated, not solved.
9. Conclusion
The framework compresses to a single line:
I do not know — but I can commit. I commit — but I leave room for being wrong.
It does not deliver truth. It does not guarantee returns. It guarantees one thing: that when the investor is wrong — and he will be wrong — he is still standing, still holding capital, still able to go again.
The methodological stance is this. Every parameter in a quantitative framework is an unverifiable belief, priced. The value of the exercise is not that it eliminates judgment — it cannot; designing is judging — but that it renders judgment explicit, decomposable by layer, and auditable against its own falsifiability horizon.
And where that horizon recedes beyond a human lifetime, the only honest response is not a better argument. It is a reserve.
References
[1] Constantinides, G. M. (1979). A Note on the Suboptimality of Dollar-Cost Averaging as an Investment Policy. Journal of Financial and Quantitative Analysis, 14(2), 443–450.
[2] Edleson, M. E. (1993). Value Averaging: The Safe and Easy Strategy for Higher Investment Returns. (Concept first published 1988; reissued 2007, John Wiley & Sons.)
[3] Bitcoin halving record (block subsidy and dates): 1st 2012-11-28 (50→25 BTC), 2nd 2016-07-09 (25→12.5), 3rd 2020-05-11 (12.5→6.25), 4th 2024-04-19 at block 840,000 (6.25→3.125); 5th projected ~2028 at block 1,050,000 (→1.5625). Halving every 210,000 blocks; supply capped at 21 million BTC. (Compiled from public on-chain records and exchange educational sources.)
[4] Meynkhard, A. (2019). Fair Market Value of Bitcoin: Halving Effect. Investment Management and Financial Innovations, 16(4), 72–85.
[5] An Empirical Examination of Bitcoin’s Halving Effects (2024). Journal of Risk and Financial Management, 17(6), 229.
[6] Synthesis of Bitcoin price-prediction research (2026): out-of-sample tests indicate the Stock-to-Flow model does not outperform a naïve current-price forecast at 1–6 month horizons.
[7] Industry analyses (2024–2025) on whether the four-year cycle is altered by spot-ETF inflows and institutional adoption following the January 2024 ETF launch and the April 2024 halving.
[8] Brennan, M. J., Li, F., & Torous, W. N. (2005). Dollar Cost Averaging. Review of Finance, 9(4), 509–535.
[9] Milevsky, M. A., & Posner, S. E. (1999/2003). A Continuous-Time Re-examination of the Inefficiency of Dollar-Cost Averaging. International Journal of Theoretical and Applied Finance.
[10] Statman, M. (1995). A Behavioral Framework for Dollar-Cost Averaging. Journal of Portfolio Management, 22(1), 70–78.
[11] Wyckoff, R. D. (1930s). The Wyckoff Method of accumulation/distribution schematics. (As systematised in subsequent technical-analysis literature; e.g. Pruden, H. O., The Three Skills of Top Trading.)
Note: the sources above are real academic and industry references supporting the paper’s methodological argument. Some volume/page details follow common citation practice; readers requiring exact editions should verify against the primary databases (JFQA, SSRN, RePEc). No source has been extended beyond its original claims.
This is a methodological working paper. It does not constitute investment advice. Crypto assets are extremely volatile and total loss of principal is possible.