Continuous Prediction Markets vs. Traditional Derivatives

TL;DR

• Traditional derivatives (futures, options, perps) are useful for short-term speculation, but weak at forecasting across time horizons or expressing full market beliefs
• Continuous prediction markets allow bets on any price at any future time, aggregating into a full probability distribution rather than a single price point
• AI agents and humans trade together: AI covers short-term precision, humans contribute broader insights, producing more accurate and adaptive forecasts
• Unified, continuous design avoids liquidity fragmentation and incorporates new information faster than siloed derivative markets
• Liquidity-weighted probability maps turn speculation into a collective, interpretable forecast of price trajectories
• Torch is one implementation: a decentralized AI-powered prediction system for crypto price discovery

Derivatives aren’t designed for forecasting

Crypto traders and analysts often rely on futures, options, and perpetual swaps to extract market expectations. But these instruments are built for trading and hedging, not forecasting. At best, they offer point estimates for short-term horizons or fragmented insights about volatility.

Futures and perps reflect short-term sentiment via funding rates or price drifts. Options can imply volatility over specific intervals. But to forecast where a token’s price might go next week, next month, or across multiple scenarios, these tools fall short. They’re siloed, expiry-bound, and liquidity-constrained.

These instruments are also inherently event-based rather than continuous: a futures contract has a fixed settlement date, an option expires at a specific moment. Forecasting, on the other hand, is continuous: it evolves over time and requires markets that can express probability across a range of outcomes and time horizons.

We need a new forecasting infrastructure, the one that treats prediction itself as the core function, not a byproduct.

A comparison snapshot

Here’s how traditional financial instruments compare to continuous prediction markets in their ability to express and extract future price expectations:

A better forecasting system

A better forecasting system has a continuous, expressive, and information-rich market design.

Forecasting across all time horizons

Continuous prediction markets allow traders and AI agents to stake on price intervals at any future time. This enables the market to form beliefs about:

• Immediate price moves (e.g. next 15 minutes)
• Medium-term events (e.g. daily/weekend volatility)
• Long-term scenarios (e.g. token repricing post-upgrade)

Because there are no hard boundaries between these horizons, participants naturally contribute where they have edge. AI agents bet on short-term data-driven signals, humans on macro, news, and fundamental trends.

Full probability distributions, not single outcomes

Instead of betting on binary outcomes or inferring expectations from derivative prices, continuous markets express belief as a distribution over price × time. The system surfaces not only what traders think will happen, but how confident they are.

The output is a probability heatmap with each (x,y) contains the aggregated probability for that outcome. This enables a richer understanding of expected volatility, skew, tail risk, and consensus beliefs, collecting them all in one place, without needing complex modeling assumptions.

Unified and efficient market structure

Traditional derivatives require many instruments to express different outcomes (e.g. one future per expiry, dozens of options for each strike and tenor). This fragments liquidity and attention.

In contrast, a continuous market aggregates all predictions into a shared surface. Every bet feeds into the overall forecast, regardless of horizon or interval. The design is similar to how AMMs concentrate liquidity in DeFi: it makes the market more resilient, adaptive, and cost-efficient.

AI and human synergy

In these markets, autonomous AI agents operate continuously, placing bets on narrow intervals where confidence is high. Humans, on the other hand, often express broader uncertainty or respond to off-chain narratives.

This creates a hybrid intelligence layer:

• AI bets sharpen the distribution, especially in the near term
• Human bets fill in broader tails and react to non-quant data

The combined result is a forecast that benefits from both signal processing and subjective foresight.

Real-time updates and incentive-aligned signals

Prediction markets only work if incentives reward early and accurate insights. In continuous systems, bets are scored dynamically:

• The earlier a correct prediction is made, the higher the reward
• Narrower (bolder) intervals earn more if accurate
• Market mechanisms (like bonding curves or liquidity curves) enforce a cost to placing low-confidence noise

This filters out random speculation and encourages information-rich forecasting, rather than passive betting.

Real-world parallels and precedents

Decentralized prediction markets

Polymarket, Omen, and other platforms demonstrated the potential of decentralized information markets. But their structure was limited:

• Each market was a discrete event
• Binary or categorical outcomes only
• Liquidity split across questions

Continuous systems build on their infrastructure (like conditional tokens) but offer more expressive, unified forecasting.

Dynamic pari-mutuel systems

Unihedge introduced the idea of infinite buy-in liquidity and incentives for early information incorporation. Dynamic pari-mutuel mechanisms like those explored by Pennock allow bettors to influence payout curves as they place bets, which is perfect for evolving distributions.

Distribution markets

Dave White’s research formalized prediction as a continuous probability distribution, updated via trading. It proposes AMMs that maintain and expose forecast curves, not prices. This concept is central to the idea of forecast-as-output rather than forecast-as-implied.

Agentic economies

Olas Predict showcases the power of AI agents in forecasting markets. Its Proof-of-Active-Agent system coordinates hundreds of autonomous traders that learn by competing on information markets. These bots:

• Trade continuously
• Specialize in different niches (e.g. data feeds, NLP sentiment)
• Collaborate and arbitrate with other bots

This real-world case proves that AI can be a primary forecasting actor, not just a signal provider.

InfoFi forecasting

Torch is a decentralized prediction system designed to forecast token prices using continuous markets. Key features:

• Continuous markets across price × time
• Range-based betting with variable sharpness and time horizons
• Liquidity-weighted probability map as public good
• AI-agent integration from day one (via Olas ecosystem)

Torch outputs a living, real-time signal of where prices are expected to go, and how likely each outcome is. It’s not a bet on a yes/no event, but a confidence surface built by traders and bots.

By aligning incentives with prediction quality (not popularity or volume), Torch redefines what a market can be: not just a trading venue, but a collective forecast engine.

From markets of risk to markets of insight

Most financial markets are designed to transfer risk. Prediction markets (especially continuous ones) are designed to surface information.

As AI agents, DeFi traders, and researchers look for ways to anticipate where crypto is headed, the tools they need are not just faster trading platforms, but smarter forecasting systems.

Continuous prediction markets offer:

• Unified price-time intelligence
• Real-time, expressive forecasts
• Open participation by humans and AI

They are the next step in the evolution of markets. Not just places to profit, but places to learn. And as platforms like Torch show, this future is already under construction.