Common Questions

An LSTM (Long Short-Term Memory) is a type of recurrent neural network designed to remember information over long sequences. Unlike basic models that forget older patterns, LSTMs have memory cells that decide what to keep and what to discard, making them naturally suited for time series data like currency movements where yesterday's price influences tomorrow's prediction.

For currency forecasting, we typically work with at least 2–3 years of daily price data to capture seasonal patterns and market cycles. You'll want to reserve roughly 70% for training, 15% for validation, and 15% for testing. The exact amount depends on your forecast horizon and how specific your target currency pair is.

No. Currency markets are influenced by countless unpredictable factors—geopolitical events, central bank announcements, economic shocks. What LSTM models do well is identify recurring patterns and give you probabilistic forecasts with measurable error margins. We focus on building models that beat naive baselines and provide actionable signals, not perfect predictions.

We use several techniques depending on the gaps. For small missing values, we apply forward-fill or interpolation. For longer gaps (like weekends or holidays), we either exclude them from the sequence or use time-aware encoding to help the model understand the break. The key is keeping the data clean before it reaches the LSTM—garbage in, garbage out applies here.

Short-term forecasts (minutes to hours) can capture intraday volatility and order flow patterns. Long-term forecasts (weeks to months) require different data features—macroeconomic indicators, interest rates, sentiment—and usually larger prediction windows. We'll advise you on what's realistic based on your timeframe and available data.

Yes. We have hands-on experience building LSTM models for CAD pairs, particularly USD/CAD and other Montreal-relevant currency movements. We understand the local economic context and which data sources are most reliable for Canadian market analysis.