Patent Races and Innovation Timing

Why Patent Races Change Research Incentives

When several firms pursue the same innovation, the first successful researcher may capture a patent prize while later research produces no prize from that particular opportunity. Each firm therefore weighs the value of winning against the chance that a rival succeeds first. The timing of success matters: a larger hazard of success means a firm is more likely to win, but greater competition can also make the opportunity disappear sooner for everyone.

A patent-race model represents research success as a probabilistic arrival process. It is a deliberately focused model of research incentives, not a complete description of innovation. In particular, its results depend on assumptions about how hazards relate to effort, when research stops, and what a patent is worth. We will state those assumptions before using the model.

Definition: A firm’s research-success hazard, denoted $\lambda_i$, is its instantaneous success rate per unit of time, conditional on no firm having succeeded yet. Under a constant hazard $\lambda_i$, the time to firm $i$’s success is exponentially distributed, with survival probability $P(T_i>t)=e^{-\lambda_i t}$ for $t\geq 0$. A greater hazard means success is more likely to arrive sooner; it is not itself a probability.

Suppose two firms conduct independent research with constant hazards $\lambda_A$ and $\lambda_B$. Assume both stop researching as soon as either firm succeeds, success immediately awards the patent to the successful firm, and simultaneous success has probability zero. Independence and constant hazards imply that the time until the first success is exponential with hazard $\lambda_A+\lambda_B$. The probability firm A is first is $\lambda_A/(\lambda_A+\lambda_B)$, provided the total hazard is positive.

Worked Example: First Success and Innovation Timing

Two hypothetical materials firms independently investigate a coating. Firm A’s constant success hazard is $0.30$ per year, and firm B’s is $0.20$ per year. Research stops when either firm succeeds. Find the probability A succeeds first, the expected time until the first success, and the probability that an innovation occurs within two years.

Method and assumptions. Use independent exponential success times with constant hazards. The total hazard is $\lambda_A+\lambda_B=0.30+0.20=0.50$ per year.

The probability A is first is $0.30/0.50=0.60$. The first-success time is exponential with rate $0.50$ per year, so its expected value is $1/0.50=2$ years.

By two years, the probability of at least one success is $1-e^{-(0.50)(2)}=1-e^{-1}\approx 0.6321$. For a further check, A’s probability of being the first success within two years is $(0.30/0.50)(1-e^{-1})\approx 0.3793$, which is less than the probability $0.60$ that A eventually succeeds first.

Check. The two firms’ probabilities of being first are $0.60$ and $0.20/0.50=0.40$, which sum to 1. A common mistake is to treat a hazard such as $0.30$ per year as a 30% probability of success within a year. The probability instead depends on the time interval: for A alone over one year it is $1-e^{-0.30}$, not $0.30$.

Expected Patent Value in a Race

To compare research choices, a firm needs the expected value of winning and the expected cost of research. Consider firm $i$ facing rivals whose combined success hazard is $\mu$. Assume the firm chooses a constant research effort $e$, giving it a constant hazard $\lambda(e)$. If it wins, it receives a fixed patent value $V$ at the instant of success. Let $r>0$ be the continuous discount rate per unit of time, and let $c(e)$ be the firm’s research cost per unit of time while the race remains active. All firms stop when the first success occurs. Costs are paid continuously until that time, whether the winner is the firm or a rival.

Under these assumptions, firm $i$’s expected discounted net payoff is

U(e;\mu)=\frac{\lambda(e)V-c(e)}{r+\lambda(e)+\mu}.

Here $V$ is a prize value at the success date, not an annual flow. The costs $c(e)$ and the product $\lambda(e)V$ are both measured per unit of time, so the payoff has the same units as $V$. If the firm’s research continues after a rival succeeds, if success hazards vary over time, or if the patent prize depends on the identity or timing of the winner, this particular expression must be changed.

Why the Payoff Formula Holds

Let $T$ be the time of the first success by any firm. With constant independent hazards, $T$ has exponential rate $\lambda(e)+\mu$, so the probability the race remains active at time $t$ is $e^{-(\lambda(e)+\mu)t}$. Firm $i$’s instantaneous chance of winning in a short interval after $t$, conditional on no earlier success, is approximately $\lambda(e)\,dt$. Discounting the prize by $e^{-rt}$ and integrating gives its expected discounted prize. The firm pays its research cost only while the race is active, so the same survival and discount factors determine expected cost:

\begin{aligned} E[\text{discounted prize}] &=\int_0^\infty e^{-rt}V\lambda(e)e^{-(\lambda(e)+\mu)t}\,dt =\frac{\lambda(e)V}{r+\lambda(e)+\mu},\\ E[\text{discounted research cost}] &=\int_0^\infty e^{-rt}c(e)e^{-(\lambda(e)+\mu)t}\,dt =\frac{c(e)}{r+\lambda(e)+\mu},\\ U(e;\mu) &=E[\text{discounted prize}]-E[\text{discounted research cost}] =\frac{\lambda(e)V-c(e)}{r+\lambda(e)+\mu}. \end{aligned}

The integral $\int_0^\infty e^{-at}\,dt=1/a$ for $a>0$ produces the denominators. This derivation also shows why the rival hazard belongs in the denominator: a rival’s success can end the period during which this firm might win and incur research costs. It is not correct to omit rivals merely because the firm’s own research hazard is the object of interest.

Worked Example: Marginal Value of Raising the Success Hazard

A hypothetical firm’s success hazard is $\lambda(e)=3e$ per year, where $e$ is effort. The patent is worth $V=\$12.10$ million at success. The discount rate is $r=0.20$ per year, and rivals have a combined hazard $\mu=1.00$ per year. At $e=1/3$, calculate the marginal increase in the firm’s expected discounted gross prize from a unit increase in effort. Ignore research cost for this calculation.

Method. The gross prize value is $G(e)=\lambda(e)V/[r+\lambda(e)+\mu]$. Differentiating with respect to effort gives $\frac{dG}{de}=\lambda'(e)V(r+\mu)/[r+\lambda(e)+\mu]^2$. At $e=1/3$, the firm’s hazard is $\lambda=3(1/3)=1.00$ per year and $\lambda'(e)=3$. The total rate in the denominator is $0.20+1.00+1.00=2.20$ per year.

Substitution gives \[ \frac{dG}{de} =\frac{3(12.10)(0.20+1.00)}{(0.20+1.00+1.00)^2} =\frac{43.56}{4.84} =9.00. \] Thus the marginal gross prize value is $\$9.00$ million per unit of effort. As a check on the levels, the gross prize value at the stated effort is $(1.00)(\$12.10\text{ million})/2.20=\$5.50$ million.

Check and common mistake. The derivative is 9, not 3: $43.56/4.84=9.00$. Reporting 3 is an arithmetic error. Another error is to omit the rival hazard from the squared denominator; the correct denominator is $(r+\lambda+\mu)^2=(2.20)^2$, not a denominator based on the firm’s own hazard alone. This is a gross-prize derivative, not the net marginal payoff after accounting for the effort cost.

Effort Choice and the Effect of Rivalry

To find an effort choice, specify how effort changes the success hazard and how costly it is. Use the simple functional forms $\lambda(e)=\alpha e$, where $\alpha>0$, and $c(e)=\kappa e^2/2$, where $\kappa>0$. Effort is nonnegative. These assumptions mean effort raises the hazard at a constant marginal rate, while the flow cost is convex. They are modelling choices, not universal features of research.

Differentiating the payoff with respect to $e$, an interior optimum satisfies

\frac{\partial U}{\partial e} = \frac{(r+\mu)\bigl(\alpha V-\kappa e\bigr)-\alpha\kappa e^2/2} {(r+\mu+\alpha e)^2} =0.

The denominator is positive. The numerator balances the marginal research benefit against the cost of effort and the fact that effort also changes the expected duration of the race. A valid interior solution must have $e>0$; the firm must also compare it with the boundary choice $e=0$. With these forms, the derivative numerator decreases through its positive root, so the root is a maximum when it lies in the feasible range. A larger prize $V$ raises the benefit of winning and, other things equal, supports more effort. A rival hazard $\mu$ changes the payoff and the optimal response through the probability and timing of winning; it is not a cost term that can simply be added to $c(e)$.

Worked Example: Optimal Effort Against a Given Rival Hazard

A hypothetical firm chooses effort $e\geq 0$. Its success hazard is $\lambda(e)=2e$ per year, and its flow research cost is $c(e)=e^2$ million dollars per year. The patent prize is $V=\$2$ million. The continuous discount rate is $r=0.20$ per year, and rivals’ combined hazard is $\mu=0.80$ per year. Find the firm’s optimal effort and expected net payoff.

Method. Apply the constant-hazard payoff formula. Since $r+\mu=1.00$ per year, the firm’s payoff, in millions of dollars, is $U(e)=(4e-e^2)/(1+2e)$. For $e\geq 0$, the denominator is positive. Differentiating, \[ U'(e)=\frac{(4-2e)(1+2e)-2(4e-e^2)}{(1+2e)^2} =\frac{4-2e-2e^2}{(1+2e)^2}. \] Setting the numerator equal to zero gives $4-2e-2e^2=0$, or $e^2+e-2=0$, so $e=1$ or $e=-2$. Only $e=1$ is feasible.

For $0\leq e<1$, $4-2e-2e^2>0$, while for $e>1$ it is negative. Thus payoff rises up to $e=1$ and falls afterward; the boundary $e=0$ has payoff zero, confirming $e=1$ is the global maximum. At that effort, the firm’s hazard is $2$ per year, the total hazard is $2+0.80=2.80$ per year, and $U(1)=(4-1)/(1+2)=\$1.00$ million.

Check and common mistake. The chosen effort produces positive net payoff, and the derivative changes from positive to negative at the chosen value. A common error is to use the rival hazard $0.80$ as the total hazard in the denominator. The correct denominator includes the firm’s own hazard as well as the discount rate and rival hazard.

Worked Example: More Rival Research and the Firm’s Response

Keep the firm’s technology, costs, prize, and discount rate from the previous example, but suppose rivals’ combined hazard rises from $0.80$ to $1.80$ per year. Find the firm’s new optimal effort and payoff against this fixed rival hazard. Compare the result with the earlier choice.

Method. Now $r+\mu=0.20+1.80=2.00$ per year, so the firm maximizes $U(e)=(4e-e^2)/(2+2e)$, in millions of dollars. Its derivative is \[ U'(e) =\frac{(4-2e)(2+2e)-2(4e-e^2)}{(2+2e)^2} =\frac{8-4e-2e^2}{(2+2e)^2}. \] The first-order condition reduces to $e^2+2e-4=0$. The feasible root is $e=-1+\sqrt{5}\approx1.2361$; the other root is negative. The derivative is positive below this root and negative above it, so this is the maximum.

At $e=-1+\sqrt{5}$, the numerator of the payoff is $4e-e^2\approx3.4164$, and the denominator is $2+2e\approx4.4721$. The payoff is therefore approximately $\$0.7639$ million. Relative to the earlier case, the firm raises effort from $1$ to about $1.2361$, but its expected payoff falls from $\$1.00$ million to about $\$0.7639$ million.

Check and common mistake. Substituting the rounded effort back into the first-order condition gives approximately $1.5279+2(1.2361)-4=0$, as required. Here more rival research increases this firm’s best response effort but reduces its payoff. The effort response is specific to the stated model and parameters; it should not be asserted as a universal law. Do not confuse a firm’s best response against a fixed rival hazard with the outcome of a model in which all firms choose effort simultaneously.

Rivalry, Duplication, and the Limits of the Model

A patent race creates two distinct effects. First, rival research can raise the total arrival rate of innovation: with independent constant hazards, the first success arrives at rate equal to the sum of the firms’ hazards. Second, the same rival activity lowers any one firm’s chance of being first. That makes research a race for a private prize rather than a joint project whose participants all receive the same reward.

If multiple firms investigate the same idea, their efforts may be partly duplicative: a rival’s research can end the opportunity for this firm even though the rival has not used this firm’s research. The model captures this in the stopping rule and the rival hazard. It does not, on its own, quantify the social value of knowledge spillovers, calculate socially optimal research, or determine how patent policy should be designed. Those questions require additional assumptions about benefits to consumers, spillovers, and the value of later innovations.

The payoff formula also relies on the prize and hazards being well-defined. A patent value that depends on market size, licensing, or the date of success may need to be represented as a time-dependent prize. If firms can continue research after a rival’s success, expected costs no longer stop at the first arrival. If hazards are not constant, probabilities and present values generally require time-dependent hazard functions rather than a single exponential distribution.

Key takeaway: In a constant-hazard race, the first-success hazard is the sum of the firms’ hazards, while a firm’s probability of winning depends on its own hazard relative to that total. The firm’s expected discounted net payoff is its expected prize less research costs, with both discounted over the period before the race ends.

Check Your Understanding

For each problem, use the stated assumptions and keep hazards, time, and discount rates in consistent units.

Two firms have independent constant hazards of $0.12$ and $0.18$ per year. Find the probability the second firm succeeds first and the expected time to first success. (Answer: $0.18/0.30=0.6000$; $1/0.30=3.3333$ years.)
Firm A and its rival have hazards $0.40$ and $0.10$ per year. Find the probability that an innovation occurs within three years and the probability A is first and succeeds within those three years. (Answer: $1-e^{-1.5}=0.7769$; $(0.40/0.50)(1-e^{-1.5})=0.6215$.)
A firm has $\lambda(e)=4e$ per year, a patent prize of $\$5$ million, discount rate $r=0.25$ per year, and rival hazard $\mu=0.75$ per year. At $e=0.25$, what is the marginal gross prize value per unit of effort? (Answer: $4(5)(0.25+0.75)/(0.25+1.00+0.75)^2=\$5.00$ million per unit of effort.)
A firm has $\lambda(e)=1.5e$ per year, prize $V=\$3$ million, flow cost $c(e)=e^2$ million dollars per year, and $r+\mu=1.5$ per year. Its payoff is $(4.5e-e^2)/(1.5+1.5e)$. Find its nonnegative effort that maximizes payoff. (Answer: $e=-1+\sqrt{5.5}\approx1.3452$.)
Two firms each have a constant success hazard of $0.25$ per year. Find the probability that an innovation occurs within one year and the probability that a specified firm is first within that year. (Answer: $1-e^{-0.50}=0.3935$; $(0.25/0.50)(1-e^{-0.50})=0.1967$.)

What You'll Learn