KongSwap prediction markets have been live now for over one month. While the reception has been very positive, after running for one month we see one major area where the current model needs to be improved - making predictions across the lifecycle of the market. We propose a Time-Weighted Pool-Based Prediction Markets: Rewarding Early Predictors Without Penalizing Accuracy.
Problem with current Pool-Based Prediction Markets on KongSwap
Our current pool-based prediction system:
- User contribute to predictions on outcomes.
- At event resolution, the total pool is distributed among accurate predictors, proportional to their share of the total correct-side contributions.
Issue:
- Users who join late but are still accurate can dilute the reward distribution.
- There’s no incentive to predict early.
We recognise the need for a better mechanism that rewards early predictions without punishing correctness.
Solution: Exponential Weighting with Return Floor
We are proposing to upgrade our pool-based prediction markets model with Time-weighted predictions, where users who make predictions earlier, get more favorable payout multipliers, to incentivize early liquidity and risk-taking.
This would result in:
- Encourages early predictions.
- Smoothly reduces reward weight over time.
- Doesn’t overly penalize latecomers.
- Correct predictors should never lose money, regardless of how late they place the prediction.
- Is fair and transparent.
Giving higher weight to earlier predictions would affect how much of the pool a participant receives if their prediction is correct.
Weighting Function
We use exponential decay to define the weight:
w(t) = ⍺^(t/T)
Where:
- ( t ) = time of the prediction (relative to market open)
- ( T ) = total duration of the market
- ( ⍺ in (0, 1) ) controls how fast the weight decays
The chart below illustrates how different values of ⍺ affect the exponential decay of time-based weights. Lower values like ⍺ = 0.1 create a steep curve that heavily rewards early predictors, while higher values like ⍺ = 0.9 produce a much gentler decay—giving late predictors more parity. We propose to do a very fast decay that will heavily favor early prediction participation.
In this version, we use ( ⍺ = 0.1 ) to extremely favor early predictors.
Example Weights (( ⍺ = 0.1 ))
| Time ( t ) | Weight ( w(t) ) |
|---|---|
| 0 | 1.000 |
| 0.5T | 0.316 |
| 0.9T | 0.126 |
Reward Mechanism With Return Floor
We split the total pool into two parts:
- Guaranteed Return (G): Ensures all correct predictors get at least their original contribution back.
- Bonus Pool (B): Remaining pool is distributed based on weighted shares.
Let:
- a_i : Amount participant ( i ) contributed
- w(t_i) : Weight at time of prediction
- c_i = a_i * w(t_i) : Weighted contribution
- W = ∑ a_i : Total correct-side contributions
- C = ∑ c_i : Total weighted correct-side contributions
- P : Total pool size
Final Reward Formula:
reward_i = a_i + (c_i/ C ) * (P - W)
This ensures:
- a_i : base guaranteed return
- (P - W) : bonus pool (from incorrect predictions)
- c_i / C : share of the bonus pool, based on time-weighted contribution
Example
Assume:
- Total pool ( P = 450 ) USDC
- 3 accurate predictors:
- Alice: 100 USDC at ( t = 0 )
- Bob: 100 USDC at ( t = 0.5T )
- Carol: 100 USDC at ( t = 0.9T )
- ( ⍺ = 0.1 )
Weights:
- Alice: ( w = 1.000, c = 100.0 )
- Bob: ( w = 0.316, c = 31.6 )
- Carol: ( w = 0.126, c = 12.6 )
Totals:
- W = 300, C = 144.2, G = 300, B = 150
Rewards:
| Participant | Contribution | Weighted ( c_i ) | Bonus Share | Final Reward |
|---|---|---|---|---|
| Alice | 100 | 100.0 | 104.0 | 204.0 |
| Bob | 100 | 31.6 | 32.9 | 132.9 |
| Carol | 100 | 12.6 | 13.1 | 113.1 |
All correct predictors receive at least their contribution back. With ( ⍺ = 0.1 ), early predictions are overwhelmingly favored in the bonus distribution. As you can see from the example the Alice would receive almost double the payout on the same prediction amount because she made her prediction at the beginning of the market as opposed to Carol who waited closer to market expiration.
Benefits of the of the Proposed Upgrade
This post introduces a better approach to pool-based prediction markets: Exponential Weighting with a Return Floor. It’s simple, mathematically robust, and aligns incentives for both early movers and accurate predictors:
- Encourages Early Participation: First movers are strongly rewarded
- Fair to All Accurate Predictors: correct predictors should never receive less than they contributed
- Smoothly reduces reward weight over time.
- Correct predictors should never lose money, regardless of how late they place the prediction.
We will be implementing this in the soon in the main prediction markets code. All markets that were created pre-upgrade will remain on the initial model whilst new markets will have a new time weighted approach to predictions.