Paper: arXiv 2406.18206

Authors: Kamil Kashif, Robert Ślepaczuk

Abstract

This study focuses on building an algorithmic investment strategy employing a hybrid approach that combines LSTM and ARIMA models referred to as LSTM-ARIMA. This unique algorithm uses LSTM to produce final predictions but boosts the results of this RNN by adding the residuals obtained from ARIMA predictions among other inputs. The algorithm is tested across three equity indices (S&P 500, FTSE 100, and CAC 40) using daily frequency data from January 2000 to August 2023. The testing architecture is based on the walk-forward procedure for the hyperparameter tunning phase that uses Random Search and backtesting the algorithms. The selection of the optimal model is determined based on adequately selected performance metrics focused on risk-adjusted return measures. We considered two strategies for each algorithm: Long-Only and Long-Short to present the situation of two various groups of investors with different investment policy restrictions. For each strategy and equity index, we compute the performance metrics and visualize the equity curve to identify the best strategy with the highest modified information ratio. The findings conclude that the LSTM-ARIMA algorithm outperforms all the other algorithms across all the equity indices which confirms the strong potential behind hybrid ML-TS (machine learning - time series) models in searching for the optimal algorithmic investment strategies.

Complexity vs Empirical Score

  • Math Complexity: 7.0/10
  • Empirical Rigor: 6.0/10 (capped from 8.5: transaction costs only partially modelled; see penalty flags)
  • Quadrant: Holy Grail — high math complexity, high empirical rigor (effective empirical rigor capped from 8.5)

Why this score: The paper employs advanced hybrid modeling techniques (LSTM-ARIMA) and rigorous statistical backtesting procedures (walk-forward optimization, random search) across multiple indices with comprehensive performance metrics, striking a balance between sophisticated mathematics and high empirical standards.

Research Flowchart

  flowchart TD
  A["Research Goal<br>Develop & test hybrid LSTM-ARIMA<br>for algorithmic investing"] --> B["Data & Models"]
  B --> C["Hybrid LSTM-ARIMA<br>combines LSTM predictions<br>with ARIMA residuals"]
  B --> D["Benchmarks<br>Standalone LSTM, ARIMA,<br>Random Walk"]
  C & D --> E["Backtesting Engine<br>Walk-forward validation<br>Hyperparameter tuning via Random Search"]
  E --> F["Strategies<br>Long-Only vs Long-Short<br>applied to S&P 500, FTSE 100, CAC 40"]
  F --> G["Outcomes<br>LSTM-ARIMA outperforms benchmarks<br>Highest Modified Information Ratio<br>Validates Hybrid ML-TS potential"]

Evidence Card

What the authors claim, on what data, and who has checked it. Read from the paper's own text by a language model: every quote is verbatim, nothing was re-run. How to read this card

Proposes

The paper proposes a hybrid algorithmic investment strategy that combines LSTM and ARIMA models, called LSTM-ARIMA, to predict next‑day closing prices.

Source sentence
This study focuses on building an algorithmic investment strategy employing a hybrid approach that combines LSTM and ARIMA models referred to as LSTM-ARIMA.

Assumes 4

  • The walk‑forward optimization reduces over‑fitting risk. model
    Source sentence
    we use walk-forward optimization (WFO) as this technique reduces the risk of over-fitting to one specific sample of past returns.
  • Predictions are made for the next day’s closing price using daily data. data
    Source sentence
    The algorithm predicted the next day’s closing price based on the historical data and was classified as a regression problem.
  • ARIMA order d is set to 1 when modeling prices. model
    Source sentence
    In terms of stock forecasting, order d is usually set to 1 when we model the prices
  • The in‑sample window is 1250 trading days (1000 training, 250 validation). execution
    Source sentence
    We set the in-sample (IS) window to 1250 trading days where the training set is equal to 1000 trading and the validation set is equal to 250 trading days

Data

Backtest sample: 2005-01 – 2023-08 · S&P 500 stocks · equities · daily

  • S&P 500 · Yahoo Finance market-prices public
    2000-01-03 to 2023-08-30
    Source sentence
    The testing architecture is based on the walk-forward procedure
  • CAC 40 · Yahoo Finance market-prices public
    2000-01-03 to 2023-08-30
    Source sentence
    the ... CAC 40 equity index on 2004-12-28 and lasts until 2023-08-30.
  • Yahoo Finance via yfinance API · Yahoo Finance market-prices public
    Source sentence
    We consider three equity indices in our research and the data is taken from Yahoo Finance using their yfinance API for Python.

Recognised sources: Yahoo Finance Cboe

Costs

Partial Costs appear only as a robustness check, a flat fee without spread or impact, or outside the headline numbers.

Finds 3

Reported performance (LSTM-ARIMA): annual return 8.9% · max drawdown 56.6%

  • LSTM-ARIMA outperforms all other models across all three equity indices according to the Modified Information Ratio. performance
    Source sentence
    The findings conclude that the LSTM-ARIMA algorithm outperforms all the other algorithms across all the equity indices which confirms the strong potential behind hybrid ML-TS
  • For FTSE 100 both Long‑Only and Long‑Short, LSTM-ARIMA achieves the highest IR** values, indicating best performance. performance
    Source sentence
    Based on the evaluation of IR**metrics for both the Long-Only and Long-Short strategies, it can be deduced that the LSTM-ARIMA algorithm outperformed the other algorithms for the FTSE 100 equity index.
  • The hybrid model’s performance varies across assets, indicating lack of robustness to asset changes. negative
    Source sentence
    Our hybrid model outperforms all the other models across all equity indices, though with varying performance metrics. The varying performance metrics across different assets indicate that our algorithms are not robust to changes in the asset.

Checked by others

No other paper in the archive has been found to support or contradict these findings (compared with 15 related papers). Coverage is partial, so this is neither replication nor refutation.

Cited by 2 papers in the archive

Code

No repository by the authors found in the paper.

Extracted 2026-11-14 · same card as JSON · A language model chose these items: check the source sentence before relying on one.