When you watch a price on your trading screen it feels immediate and exact, but the truth is that what you see can come in different forms. Some feeds deliver near-raw, tick-by-tick messages from exchanges; others return normalized or packaged data that has been transformed for convenience, bandwidth, or analytics. This article explains the difference, why vendors apply smoothing or filtering, how that changes the numbers you see, and what to check depending on your trading needs. Trading carries risk; this is educational material only and not personalized advice.
What “raw” market data means
Raw market data is the feed emitted by an exchange or consolidated tape in its native format. It contains the low-level messages that describe individual trades, quote updates, order adds/cancels (in venues that publish them), sequence numbers, and ancillary administrative messages. In practice raw feeds are often multicast UDP streams or packet captures captured at the network interface. They show the market microstructure: every trade, every quote change, and every tiny event as the exchange reports it, usually with the highest available timestamp precision.
For example, a raw equity feed might show three individual trades at 12:30:01.001, .003, and .007 for prices 100.01, 100.03 and 99.95. A snapshot of the raw order book would include the exact levels and sizes at each update. Raw feeds are large, jumpy, and require specialized software and co-location or high-bandwidth links to process reliably.
What “normalized” or “packaged” data is
Most traders — especially retail traders and smaller firms — get normalized or packaged data from vendors or brokerage platforms. Normalization standardizes different exchange formats into a single schema, reduces message volume, and may add derived fields (for example, implied volatility or greeks). Packaged data is often aggregated into convenient forms for display: candles, best bid/ask (Level I), or a reduced depth-of-book.
Normalization makes life easier: your charting library only needs one format, storage needs drop, and analytics can be attached. But the process is rarely lossless. Vendors may drop administrative messages, trim timestamp precision, map symbols to a single symbology, or coalesce very fast sequences of messages into a single tidy update.
Imagine a vendor collects feeds from several exchanges and publishes a single “consolidated” price. That consolidated quote may appear smoother than any individual raw feed because the vendor has applied rules to resolve simultaneous quotes or to represent the “best” top-of-book across venues.
Where smoothing and filtering happen — and why
There are two distinct kinds of smoothing you’ll encounter: explicit analytical smoothing and operational filtering.
Analytical smoothing is applied to produce cleaner, model-friendly values. Examples include moving averages, kernel smoothing of an implied volatility surface, spline fits for theoretical option prices, or model-free interpolation that enforces arbitrage-free shapes across strikes. These techniques make derived outputs easier to use for pricing and visualization. An options vendor might compute a smoothed implied volatility surface so traders can see a continuous curve even where strikes are illiquid.
Operational filtering is applied to reduce bandwidth, remove redundant or administrative traffic, and handle noisy or malformed messages. This includes removing heartbeats, collapsing microsecond burst updates into single updates for a display client, lower-precision timestamps, or dropping messages considered irrelevant for most users. Both types change the raw event stream.
Concrete examples to make it clear
A retail charting app that shows 1-minute candles is a classic example of “not raw.” During that minute there may have been dozens of ticks, a flash spike, and multiple bid/ask sweeps. The candle shows open/high/low/close and aggregated volume — useful for visual trading but not the same as watching every tick.
An options data vendor might publish a “theoretical price” for every strike derived from available bids and offers and smoothed across strikes so the surface is arbitrage-free. That theoretical price is useful for hedging calculations but differs from the last quoted market trade.
A market data vendor advertising “normalized real-time feed” may have dropped some low-value administrative messages, mapped prices to floating point precision instead of integer ticks, and standardized timestamps to UTC. The feed will behave correctly for most strategies, but some microstructure-sensitive algorithms will perform differently than if they used raw multicast.
How to tell what you’re receiving
Vendors usually document what they provide, but the documentation can be technical. Practical checks you can do include comparing timestamps and message rates against a second vendor, asking about sequence numbers and whether packet captures (PCAPs) are available, and verifying whether the feed includes Level II/depth-of-book and execution messages or only top-of-book trades. If you need truly raw multicast feeds or packet captures you must usually request them expressly — they are more expensive and require more infrastructure.
One simple experiment: open two sources for the same symbol — your broker’s chart and a raw-tick-capable feed, if available — and look for differences in intraday micro-moments such as flash spikes or bid-ask sweeps. Differences indicate smoothing, aggregation, or normalization is in place.
Which type you should prefer — depends on your strategy
If your system relies on millisecond-level order book dynamics (for example, market making or latency arbitrage), you need direct or near-direct exchange feeds and the infrastructure to process them. Normalized feeds usually add measurable latency and remove detail.
If you are a discretionary retail trader, a packaged feed with clean candles, top-of-book quotes, and derived analytics is often sufficient and far easier to manage. Many algorithmic strategies that operate on minute bars or higher frequency than intraday but slower than microsecond-level also work well with normalized data.
Risks and caveats
Smoothing and normalization are trade-offs: they make data easier to use and cheaper to deliver but can remove signals that matter for certain strategies. Smoothed implied volatilities or theoretical option prices can hide illiquidity or arbitrage opportunities that exist in raw quotes. Aggregation can mask brief liquidity drains or sudden order cancellations that would matter to a high-frequency strategy, leading to unexpected slippage. Vendors can also make errors in normalization — mis-mapping symbols, mishandling corporate actions, or introducing small timing shifts — and those errors are harder to spot because the raw wire format isn’t visible.
Licensing and usage rules also differ: some data is licensed for display only while non-display (algorithmic) use costs more. And while many vendors use accurate methods, no system is immune from gaps, delays, or processing bugs. Always verify claims such as “raw” or “lossless” with documentation and, where feasible, sample data. Finally, remember that good execution depends on more than feed quality: routing, broker infrastructure, and your own software all affect outcomes. Trading carries risk; this article is educational and not investment advice.
How to ask vendors the right questions
When evaluating a data vendor, ask whether their feed is a direct exchange multicast or a normalized rebroadcast. Ask what they drop during normalization, how timestamps and sequence numbers are handled, whether they offer packet captures, and what derived fields they compute (and how). Confirm whether historical data is reconstructed from normalized events or from lossless captures, and whether corporate actions and splits are adjusted in the values you receive.
Also ask about latency numbers and recovery procedures: if you miss packets, how does the vendor reconstruct state? For options and complex instruments, ask how implied volatilities and surface smooths are calculated and whether raw bid/ask messages are available for verification.
Key takeaways
- What you see on a retail platform is often normalized or packaged for display; raw multicast feeds show every tick and require special infrastructure.
- Smoothing (moving averages, surface fits, model-free interpolation) helps analysis but can change the exact numbers and hide microstructure.
- Choose raw, normalized, or packaged feeds based on your strategy: HFT needs raw; most retail and many quant strategies work with normalized feeds.
- Always verify vendor claims, check timestamps/sequence behavior, and understand licensing and recovery options before relying on a feed.
References
- https://spiderrock.net/what-is-a-market-data-feed/
- https://databento.com/microstructure/normalization
- https://spiderrock.net/options-market-data-raw-and-normalized-feeds/
- https://lime.co/news/understanding-real-time-market-data-types-sources-and-what-matters-most-to-traders-143343/
- https://kb.dxfeed.com/en/data/calculated-data/options-analytics-solution.html
- https://www.investopedia.com/terms/d/data-smoothing.asp
- https://site.financialmodelingprep.com/education/data/understanding-stock-market-data-sets-realtime-vs-historical-vs-intraday