How I stopped losing sleep over spreads
I used to stare at the daily roll and wonder why my hedge fund's treasury desk always seemed to lose three basis points to the bid-ask before the market even opened. We were retail clients using a broker that gave us "live" rates, but the feed was throttled. The numbers looked correct until you tried to execute a sizeable order, then the price slipped just enough to eat your margin. That's when I started looking at OANDA's raw data instead of whatever display their platform showed. OANDA publishes historical tick data through their Data Services division, which is separate from their retail trading platform. Most people don't know that because the link sits buried under the research section. The API endpoint gives you minute-level closes, intraday bars from one-minute up to monthly, and they cover every major pair plus crosses. The key thing nobody mentions upfront is that OANDA doesn't include quotes from the interbank overnight session. If you're backtesting a strategy that assumes the London close rolls directly into Tokyo open, your results will be wrong by design. I ran into this in 2019 when I was optimizing a carry trade overlay for a multi-strategy fund. The model showed solid positive returns from December through March, which should have been a red flag because EUR/USD carry trades in winter tend to compress, not expand. I pulled the raw OANDA data and discovered the broker's closing price was sourced from a proprietary aggregation window that excluded the quiet hours where the actual supply-demand imbalance showed up. The fix was to merge OANDA's intraday data with FIX-level snapshots from a separate liquidity provider for the 23:00 to 01:00 UTC window. Once I did that, the model's performance dropped by forty-two percent over the full sample. Not a disaster, but enough to change the allocation.
What the rates actually tell you
Exchange rate movements decompose into three pieces: spot drift, forward points, and cross-rate arb. Spot drift is what most people watch, which is why they keep getting hurt. Forward points move independently of the spot because they reflect interest rate differentials and local money market conditions. The cross-rate arb shows up when two correlated pairs diverge beyond a threshold that should never hold for more than a few minutes. I track all three in a spreadsheet that refreshes every fifteen minutes during the overlap session. The forward point component alone accounts for about sixty percent of the annualized return on most carry strategies, which means if you're ignoring it you're only trading half the instrument. Another thing that trips people up is the difference between mid-market rates and executable rates. OANDA's historical data shows mid-point calculations derived from their internal pricing engine. If you need execution-grade rates for backtesting, you have to adjust by roughly one to two pips depending on the pair. EUR/USD tends to sit closer to one pip because of depth. Exotic crosses like TRY/USD can slip three to four pips in volatility spikes. I apply a fixed spread adjustment based on historical rolling volatility, then verify against actual filled prices from my broker's execution logs. It cuts backtest-to-live slippage from about eight basis points down to two.
Practical workflow
Download the OANDA Historical Data API key from their developer portal. The free tier gives you one hundred thousand requests per month, which is plenty for daily rebalancing and monthly reporting. I keep a Python script that pulls the last ninety days of one-minute bars for the core seven majors, calculates the rolling correlation matrix, and flags any pair where the correlation breaks below zero point seven for more than three consecutive sessions. That usually signals a regime shift that requires adjusting the portfolio hedge ratio. The script also cross-checks the implied forward points against the actual swap rates published by the Federal Reserve and the ECB. When the delta exceeds ten percent, I adjust the position sizing because the market is pricing in either a central bank intervention or a liquidity crunch. This has happened twice in the last five years. Both times it was correct. If you're just starting out, don't bother with the tick-level data. The one-minute bars give you everything you need for most discretionary and quantitative approaches. Tick data becomes necessary only if you're modeling order book dynamics or high-frequency mean reversion, and that's a completely different skill set. Most retail traders confuse the two and end up spending twenty hours cleaning data that doesn't improve their edge.
Get the Full Details
OANDA also publishes a daily snapshot report that includes their quoted bid, ask, and mid for every pair. It's not as granular as the API, but it's useful for checking whether a particular date had anomalous pricing during events like the Swiss franc delinking in 2015 or the BoJ yield curve control shift in 2024. I download these manually and keep them in a separate folder labeled "regime shifts" because that's what they are.
Where it falls apart
The main limitation is that OANDA data doesn't include quotes from every broker simultaneously. Their snapshot reflects their own liquidity pool and the syndicated feeds they draw from. If you're trading with a different prime broker, your executable rates will differ, sometimes materially during stress periods. I learned this the hard way in early 2023 when SVB collapsed and the dollar index spiked three percent in twenty minutes. OANDA's feed held relatively stable because of their hedging desk, but several smaller brokers widened spreads to fifty or sixty pips and some temporarily halted trading. My positions got filled at exactly OANDA's quoted price, which was right for their book but would have been disastrous if I'd been routing orders elsewhere. The second limitation is latency. Even the real-time feed has about a two hundred millisecond delay compared to direct exchange feeds. That doesn't matter for intraday swing strategies or longer-term allocation adjustments, but it absolutely matters for scalping or arbitrage. If you're doing either of those, you need a different data source entirely, preferably a FIX connection directly to the liquidity provider.
Bottom line
OANDA's rate data is good enough for most practical purposes. It's clean, well-documented, and covers the pairs that matter. The trick is understanding what it doesn't show, adjusting for the spread, and merging it with forward point data when your strategy depends on interest rate differentials. I've been running this exact pipeline for six years now. The returns aren't spectacular, but the consistency is. And in this business, consistent beats impressive every time. One final note that people miss: OANDA stops publishing historical data for certain pairs during periods of extreme volatility without updating their documentation. I found out the hard way during the UAH depeg in March 2022. The API returned empty responses for a full forty-eight hours. I had to fall back to a third-party aggregator to fill the gap. Make sure you have a backup data source before you need it, not after.
